JUN 8 1981
1981年6月8日

PROJECT REPORT 80-25
项目报告80-25

FINAL REPORT OF THE GARYINCOME MAINTENANCE EXPERIMENT:LABOR SUPPLY
GARYINCOME维持实验的最终报告：劳动力供给

LIBRARY
图书馆

MATHEMATICA POLICY RESEARCH
MATHEMATICA 政策研究

P.0. BOX 2393
P.0. 框2393

PRINCETON, NEW JERSEY 08540
普林斯顿， 新泽西州 08540

Kenneth C. Kehrer
肯尼斯·科勒（Kenneth C.Kehrer）

Mathematica Policy Research
Mathematica 政策研究

John F. McDonaldUniversity of Illinois at Chicago Circle
John F. McDonald伊利诺伊大学芝加哥分校

Robert A. Moffitt
罗伯特·莫菲特

Rutgers University
罗格斯大学

November 1979
1979 年 11 月

CONTENTS
内容

PAGE
页

ACKNOWLEDGEMENTS.……………………………………………………………………………………………………
确认。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。

INTRODUCTION.………………………………………………………………………………………………………………… 1
介绍。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。。1

I.MODELSPECIFICATIONANDMETHODOLOGICALISSUES⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 5
一、模型规范与方法问题⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 5

A.THEGARYEXPERIMENT……………………………………………………………………………… 5
A.THEGARYEXPERIMENT..........................................................................................5

B.ANAVERAGE-TAX-RATEMODEL…………………………………………………………………… 10
B.平均税率模型..............................................................................10

C.EMPIRICALSPECIFICATTONISSUES.………………………………………………………… 16
C.实证具体问题...................................................................16

1.DependentVariable………………………………………………………………………… 17
1.DependentVariable....................................................................................17

2.BudgetLine.⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 18
2.预算线.⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 18

3.EstimatingTechnique.…………………………………………………………………… 19
3.估算技术...............................................................................19

4.SampleSelection.…………………………………………………………………………… 19
4.SampleSelection........................................................................................19

5.TimePeriodsExamined………………………………………………………………… 20
5.TimePeriodsExamined...........................................................................20

6. Sample Stratification and Preenrollment Differences . 21
6.样本分层和入组前差异。21

7.WageRate⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 22
7.工资率⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯22

8. Interdependence of Family Labor-Supply Decisions. . . 24
8. 家庭劳动力供给决策的相互依存关系. . .24

9.OtherIndependentVariables⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 26
9.其他独立变量⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 26

10.OtherIssues……………………………………………………………………………………. 26
10.其他问题.................................................................................................26

D.BASICMODELRESTATED⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 29
D.BASICMODEL重述⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 29

II.FINDINGS………………………………………………………………………………………………………………… 31
二、调查结果.................................................................................................................................31

A.BASICMODEL………………………………………………………………………………………………… 32
A.基本模型...............................................................................................................32

B.SENSITIVITYTESTS…………………………………………………………………………………… 38
B.敏感性测试................................................................................................38

1.DependentVariable………………………………………………………………………… 38
1.DependentVariable....................................................................................38

2.EstimatingTechnique.⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 38
2.估算技术.⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 38

3.IncomeUnderreporting⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 41
3.收入少报⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 41

4.SampleSelection…………………………………………………………………………… 41
4.SampleSelection.......................................................................................41

5.PreenrollmentDifferences………………………………………………………… 42
5.Preenrollment差异..................................................................42

6.TimePeriod.………………………………………………………………………………………… 43
6.时间段.......................................................................................................43

7.WageRate⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 43
7.工资率⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯43

8.HouseholdInterdependence………………………………………………………… 44
8.家庭相互依存..................................................................44

9.OtherTests.⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 44
9.其他测试.⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ 44

PAGE
页

III. SUMMARY. . . . . . . . . . . . . · · · · · · · · · · · · ·45
III. 摘要 . .· · · · · · · · · · · · ·45

APPENDIX A: Sample Cha racter ist ics……… . . 47
附录 A：样本 Cha racter ist ics......... ..47

APPENDIX B: Calculation of Tax and Transfer Amounts . . . 60
附录 B：税款和转移金额的计算 . .60

APPENDIX C: Properties of Different PreenrollmentControl Methods . . · · · · · · · · · · ·
附录 C：不同 PreenrollmentControl 方法的属性。.· · · · · · · · · · ·

APPENDIX D: Wage Predicting Equations . . . . . . . . . . 67
附录 D：工资预测方程 . .67

APPENDIX E: Adjusted-Means Equations. . . . . . . . . . . .69
附录 E：调整均值方程 . .69

APPENDIX F: Full Regression Results: Basic Model . . . . 74
附录 F：完整回归结果：基本模型 . . . .74

APPENDIX G. Estimation of A Local-Tax-Rate Model. . . . . 77
附录 G. 地方税率模型的估计 . .77

REFERENCES.⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ · · · · · · · · · · · · ·81
参考资料.⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯ · · · ·

ACKNOWLEDGEMENTS
确认

This paper reports on research conducted by Mathematic a PolicyResearch under contract HEW 100-78-59 from the U. S. Department of Health,Education, and Welfare (HEW). This study is the culmination of several years of research design, data collection, model building, and analysis tasks under three different contracts. Kenneth Kehre r was project director for all of this work.
本文报告了 Mathematic a PolicyResearch 根据美国卫生、教育和福利部 （HEW） 的 HEW 100-78-59 合同进行的研究。这项研究是在三个不同的合同下进行数年研究设计、数据收集、模型构建和分析任务的结晶。Kenneth Kehre r 是所有这些工作的项目总监。

The work began as part of the conduct of the experiment atIndiana University Northwest, under contract SRS-70-63 between HEW, theIndiana State Department of Public Welfare, and Indiana University. In addition to the authors of this report, Richard Kaluzny, Lois Shaw, andStanley Stephenson, Jr. were part of a research team that refined the data collection design, grappled with the difficult model specification issues, created analysis files and conducted the initial analysis. The results of this work were presented in the Initial Findings Report on the Gary experiment (Kehre r and Moffitt, 1976).
这项工作始于印第安纳大学西北分校进行实验的一部分，根据 HEW、印第安纳州公共福利部和印第安纳大学之间的合同 SRS-70-63。除本报告的作者外，Richard Kaluzny、Lois Shaw 和 Stanley Stephenson， Jr. 也是研究团队的一员，该团队改进了数据收集设计，解决了困难的模型规格问题，创建了分析文件并进行了初步分析。这项工作的结果在关于Gary实验的初步发现报告中（Kehre r和Moffitt，1976）中介绍。

Further model development, file construction, and analysis took place at Mathematic a Policy Research under contract HEW 100-76-73.Preliminary results of this work were presented in a number of papers at the ASSA Meetings in Atlanta City in September, 1976 (Kehrer et al.,1976; Shaw et al., 1976). This paper reports on further research on the labor supply response conducted under HEW contract 100-78-59, and constitutes the final report on the labor supply analysis of the Gary experiment. A condensed version is scheduled to appear in the Journal of Human Resources (Moffitt, 1979).
根据合同HEW 100-76-73，在Mathematic a Policy Research进行了进一步的模型开发，文件构建和分析.这项工作的初步结果在1976年9月在亚特兰大市举行的ASSA会议上的多篇论文中进行了介绍（Kehrer等人，1976;Shaw 等人，1976 年）。本文报告了在HEW合同100-78-59下进行的劳动力供给响应的进一步研究，并构成了Gary实验劳动力供给分析的最终报告。精简版计划发表在《人力资源杂志》（Moffitt，1979年）上。

Several individuals made important contributions to the development of the data. In the early stages of the work, Lois Shaw(now at the Center for Human Resource Research) had major responsibility for the development of the analysis files. The final longitudinal data files were created under the direction of Professor John McDonald at the University of Illinois at Chicago Circle. This work required the care of several other people, including Donna Vandenbrink, LaJean Waller,David Gassman, Belva Wood, Denis Ables, Adolph Hendrickson, Pal Khera,John Friedmann and Carolyn Roth.
一些人为数据的开发做出了重要贡献。在工作的早期阶段，Lois Shaw（现就职于人力资源研究中心）主要负责分析文件的开发。最终的纵向数据文件是在伊利诺伊大学芝加哥分校的John McDonald教授的指导下创建的。这项工作需要其他几个人的照顾，包括唐娜·范登布林克、拉让·沃勒、大卫·加斯曼、贝尔瓦·伍德、丹尼斯·阿布尔斯、阿道夫·亨德里克森、帕尔·赫拉、约翰·弗里德曼和卡罗琳·罗斯。

The main conceptual development of the labor-supply model was carried out by Professor Robert Moffitt at Rutgers University , who is responsible for most of the estimates presented in this report. In addition, several other individuals made important contributions to the development of the analysis over the years. These include Gary Burtless,David Greenberg, Robinson Hollister, David Horner, Stuart Kerachsky,Charles Mallar, Larry Orr, and Douglas Wolf.
劳动力供给模型的主要概念发展是由罗格斯大学的罗伯特·莫菲特教授进行的，他负责本报告中提出的大部分估计。此外，多年来，其他几个人为分析的发展做出了重要贡献。其中包括加里·伯特利斯、大卫·格林伯格、罗宾逊·霍利斯特、大卫·霍纳、斯图尔特·克拉奇斯基、查尔斯·马拉尔、拉里·奥尔和道格拉斯·沃尔夫。

Finally, we would like to thank all those who provided computational assistance for the analysis. First and foremost this includes John Friedmann, who estimated the adjusted-means regressions and developed many of the analysis files. In addition, significant contributions were made by Margo Hoft, Sandra Jamieson, Erno Krakkei,Gayla Olsen, and Debra Tessier.
最后，我们要感谢所有为分析提供计算帮助的人。首先，这包括约翰·弗里德曼（John Friedmann），他估计了调整后的均值回归并开发了许多分析文件。此外，Margo Hoft、Sandra Jamieson、Erno Krakkei、Gayla Olsen 和 Debra Tessier 也做出了重大贡献。

ii
第二

INTRODUCTION
介绍

The Gary Income Maintenance Experiment was one of a coordinated series of experiments supported by the Department of Health, Education,and Welfare and the Office of Economic Opportunity to test the work-incentive effects and other consequences of alternative negative income tax (NIT) plans. The experiments were conducted with different population groups in different parts of the country. The income-support plans tested were similar in structure to those of existing welfare and transfer programs, except that the benefit formulas were simplified and eligibility was more universal, depending only on family income, family size, and the presence of a dependent child.
加里收入维持实验是由卫生、教育和福利部和经济机会办公室支持的一系列协调实验之一，旨在测试替代性负所得税 （NIT） 计划的工作激励效应和其他后果。这些实验是在该国不同地区的不同人群中进行的。测试的收入支助计划在结构上与现有福利和转移方案的方案相似，只是福利公式得到了简化，资格更加普遍，仅取决于家庭收入、家庭规模和是否有受抚养子女。

Under an NIT, benefits are determined by a guarantee level (that is, the basic benefit provided to a family with no other source of income)and a tax rate (that is, the rate at which the benefit is reduced as other sources of income increase). Some benefits are paid to all families with incomes below a breakeven level, with the largest benefits going to those families with the lowest incomes. Thus, under such a plan, the size of the benefit decreases as family income rises, but total family income always increases as earnings from work increase.
根据国家所得税，福利由保证水平（即向没有其他收入来源的家庭提供的基本福利）和税率（即随着其他收入来源的增加而减少福利的比率）决定。一些福利金支付给所有收入低于盈亏平衡水平的家庭，其中收入最低的家庭受益最大。因此，在这种计划下，福利金的规模随着家庭收入的增加而减少，但家庭总收入总是随着工作收入的增加而增加。

While an NIT is target effective (the highest benefits go to the poorest families) and has a reasonable incentive structure (total income always rises with increments in hours worked), it nonetheless creates disincentives to work. The potential effect of an NIT on labor supply is shown in Figure l. Assume an individual facing an effective wage rateFX/FY. An NIT with a guarantee level of YG and a tax rate of t percent
虽然NIT是有针对性的（最高福利流向最贫困的家庭），并且具有合理的激励结构（总收入总是随着工作时间的增加而增加），但它仍然会抑制工作。NIT对劳动力供应的潜在影响如图l所示。假设一个人面临有效工资率FX/FY。保证级别为YG，税率为t%的NIT

FIGURE 1.
图 1.

EFFECTS OF A NEGATIVE INCOME TAX
负所得税的影响

ON LABOR SUPPLY
关于劳动力供应

Zero Hours Worked
零工作时间

would shift the individual's budget time to YGBX, where BE is the break-even level and the slope of the budget line below B is the wage rate times one minus the tax rate. The effect of an NIT would thus be to increase the income and reduce the net wage of many individuals. Both of these effects would be expected to reduce labor supply.
将个人的预算时间转移到 YGBX，其中 BE 是盈亏平衡水平，低于 B 的预算线斜率是工资率乘以 1 减去税率。因此，NIT的效果将是增加许多人的收入并减少净工资。预计这两种影响都会减少劳动力供应。

Available estimates of the potential labor supply reductions that would result from alternative NIT plans varied so widely in the1960's that they were of little use in policy planning. One of the problems underlying those estimates was that existing data did not encompass the variation in tax rates and guarantee levels that were being considered in policy planning. Perhaps more important, those estimates were based on nonexperimental data, in which observed labor supply decisions and welfare participation decisions were potentially endogenous.For example, any nonexperimental study of the labor supply of AFDC families possesses a potential selectitivity bias arising from the fact that families go on AFDC in part through a labor supply decision. A study of the labor supply behavior of both AFDC and non-AFDC families cannot easily account for the potential differences in unobserved characteristics that may have led the families to sort themselves out into the two groups. These problems have been discussed more extensively elsewhere (Moffitt andKehrer, forthcoming).
在1960年代，对替代NIT计划可能导致的潜在劳动力供应减少的现有估计差异如此之大，以至于它们在政策规划中几乎没有用处。这些估计所依据的问题之一是，现有数据没有包括政策规划中正在考虑的税率和保证水平的变化。也许更重要的是，这些估计是基于非实验数据，其中观察到的劳动力供应决策和福利参与决策可能是内生的。例如，任何关于AFDC家庭劳动力供应的非实验性研究都具有潜在的选择性偏差，因为家庭在一定程度上是通过劳动力供应决定进行AFDC的。对AFDC和非AFDC家庭的劳动力供应行为的研究不容易解释可能导致家庭将自己分为两组的未观察到的特征的潜在差异。这些问题在其他地方已经进行了更广泛的讨论（Moffitt和Kehrer，即将出版）。

In order to obtain better estimates of the likely effects on labor supply of alternative NIT plans, the Office of Economic Opportunity and HEW launched a coordinated series of income maintenance experiments that were conducted over the past decade. The income maintenance experiments generally followed a classical experimental design, wherein
为了更好地估计替代NIT计划对劳动力供应的可能影响，经济机会办公室和HEW在过去十年中启动了一系列协调的收入维持实验。收入维持实验通常遵循经典的实验设计，其中

a "treatment" is given to a randomly selected group of individuals, and their behavior is compared to that of a comparable group of individuals who do not receive the treatment. This design represents an attempt to eliminate the bias caused by unobserved variables, for the randomization that selects individuals for the experimental and control groups presumably results in identical distributions of unobserved variables within each group. Consequently, experimental data give the analyst an independent variable--treatment assignment-whose source of variation is truly exogenous.
对随机选择的一组个体进行“治疗”，并将他们的行为与未接受治疗的可比个体组的行为进行比较。这种设计代表了一种消除由未观察到的变量引起的偏差的尝试，因为为实验组和对照组选择个体的随机化可能会导致每组内未观察变量的相同分布。因此，实验数据为分析者提供了一个自变量——治疗分配——其变异来源确实是外生的。

In this paper we report on a major study of the labor supply response to the Gary experiment. Section I describes the model specifi-cations we use in the research. After a discussion of the important characteristics of the Gary experiment (Part A), we discuss the nature of the cumulative tax rate problem and derive an average-tax-rate model(Part B). In Part C we examine several other issues of empirical specification of labor supply models, and describe our approach to each of them. We restate our basic model in Part D, incorporating the specifications discussed in Part C.
在本文中，我们报告了一项关于加里实验对劳动力供给响应的重要研究。第一部分描述了我们在研究中使用的模型规格。在讨论了Gary实验的重要特征（A部分）之后，我们讨论了累积税率问题的性质，并推导了平均税率模型（B部分）。在C部分中，我们研究了劳动力供给模型的经验规范的其他几个问题，并描述了我们对每个问题的方法。我们在 D 部分重申了我们的基本模型，并纳入了 C 部分中讨论的规格。

Our findings are presented in Section II. The results of the basic model are presented in Part A and tests of the sensitivity of the results to alternative specifications decisions are presented in Part B.We summarize our findings and our interpretation of them in Section III.
我们的研究结果在第二部分介绍。基本模型的结果在A部分介绍，结果对替代规格决定的敏感性检验在B部分介绍。

4

I. MODEL SPECIFICATION AND METHODOLOGICAL ISSUES
一、模型规格和方法问题

A. THE GARY EXPERIMENT
A. 加里实验

The Gary income maintenance experiment was conducted between1971 and 1974 by Indiana University under contracts with the U. S.Department of Health, Education, and Welfare (HEW) and the Indiana StateDepartment of Public Welfare. Here we describe the design and implementation of the experiment, the characteristics of the sample and of the Gary labor market, and the interpretation and generalizability of the labor supply findings from the experiment.
1971 年至 1974 年间，印第安纳大学根据与美国卫生、教育和福利部 （HEW） 和印第安纳州公共福利部的合同进行了加里收入维持实验。在这里，我们描述了实验的设计和实施，样本和加里劳动力市场的特征，以及实验中劳动力供应结果的解释和可推广性。

Four different negative income tax plans, combining two tax rates and two guarantee levels, were tested in Gary. The tax rates were either 40 or 60 percent, and the guarantee levels were either equal to the poverty level or about three-fourths of the poverty-level annual income for each family size. In 1972, for example, when the official poverty threshold for a four-person, nonfarm family was $4,275 the twoGary guarantee levels were $4,300 and $3,300 for that family size.Benefit schedules were adjusted every six months to compensate for increases in the cost of living.
在加里测试了四种不同的负所得税计划，结合了两种税率和两种担保水平。税率为40%或60%，保证水平要么等于贫困线，要么等于每个家庭的贫困线年收入的四分之三左右。例如，在1972年，当一个四口非农家庭的官方贫困线为4,275美元时，该家庭规模的两个加里保证水平分别为4,300美元和3,300美元。福利计划每六个月调整一次，以弥补生活费用的增加。

The experiment enrolled both a husband present) and female-headed families (i. e., those without a male on the experiment because neither the New Jersey nor the Rural experiments were designed to study such families. One objective of the Gary experiment was to examine the effects of switching some female-headed families from
该实验招募了一位丈夫在场）和女性户主家庭（即那些没有男性的家庭，因为新泽西州和农村的实验都不是为了研究这样的家庭而设计的。加里实验的一个目的是研究将一些女性户主家庭从

5

AFDC to more generous income-support plans. The lower Gary guarantee level (three-fourths of the poverty level) was about $1,000 a year more than the support level of the Indiana AFDC program. Another objective of the Gary experiment was to investigate the consequences of extending eligibility for income-support payments to intact families. These families were not generally eligible for AFDC in Indiana (Indiana did not participate in the AFDC-UF program).
AFDC提供更慷慨的收入支持计划。较低的加里保证水平（贫困线的四分之三）每年比印第安纳州AFDC计划的支持水平高出约1,000美元。加里实验的另一个目的是调查将收入支持付款的资格扩大到完整家庭的后果。这些家庭通常没有资格在印第安纳州参加AFDC（印第安纳州没有参加AFDC-UF计划）。

The Gary experiment enrolled only black families, and eligibility was limited to families with at least one child under age eighteen. Of the 1,799 families who were voluntarily enrolled, 57 percent were randomly assigned eligibility for experimental income-support payments, while the remainder were control subjects. Almost 60 percent of the participating families were female-headed families.¹/
加里实验只招募了黑人家庭，资格仅限于至少有一个18岁以下孩子的家庭。在自愿入组的1,799个家庭中，57%的家庭被随机分配了实验性收入支持付款的资格，而其余的则是对照组。近60%的参与家庭是女性户主家庭。

The experimental group families were eligible for the income-support payments for three years. All participating families, both experimental and control group, filed monthly reports of income and family composition changes. All were interviewed before the experiment, about three times a year during the experiment, and after the experiment.
实验组家庭有资格获得三年的收入支助金。所有参与的家庭，包括实验组和对照组，都提交了收入和家庭组成变化的月度报告。所有人都在实验前接受采访，在实验期间和实验后每年大约三次。

The families with a male head of household present (almost all of which were intact husband-wife families) usually had low incomes but generally were not extremely poor. The husbands were typically full-time unionized workers with a history of continuous employment, who were able to earn enough to keep their families out of poverty (only 10percent of these families had incomes below the poverty line.) The wives,on the other hand, typically did not work outside the home (only 13
有男性户主的家庭（几乎所有家庭都是完整的夫妻家庭）通常收入较低，但通常不是极度贫困。丈夫通常是全职工会工人，有连续就业的历史，他们能够赚到足够的钱来让家人摆脱贫困（这些家庭中只有10%的收入低于贫困线）。另一方面，妻子通常不在家外工作（只有 13 人

1/For a more detailed discussion of the design of the experiment,see Kehrer et al. (1975).
1/有关实验设计的更详细讨论，请参阅Kehrer et al. （1975） 。

percent were employed at the start of the experiment). In the relatively few families where both the husband and wife were employed, the wife's earnings usually raised family income enough that the family no longer qualified for the receipt of NIT payments. Thus the husband-wife families studied did not include many families where the wife was working temporarily to help "make ends meet."
在实验开始时雇用的百分比）。在丈夫和妻子都有工作的家庭中，妻子的收入通常足以增加家庭收入，以至于家庭不再有资格获得NIT付款。因此，所研究的夫妻家庭并不包括许多妻子临时工作以帮助“维持生计”的家庭。

The husband-wife families studied in Gary would not be considered typical welfare families because of their attachment to the labor force and their income levels, and because public assistance payments were note generally available to husband-wife families in Indiana. But under the income-support plans tested in Gary, many of these families were eligible to receive modest income supplements. Therefore, the analysis of theGary experiment can provide ins ight into the consequences of extending an income-supplement program to working, but low-income, families. On the other hand, neither the husbands (working in unionized jobs and possessing seniority) nor the wives (either not working or working in a job that put the family income above the breakeven levels of the NIT plans) would be expected to reduce their work effort significantly in response to the experiment.
在加里学习的夫妻家庭不会被视为典型的福利家庭，因为他们对劳动力的依恋和他们的收入水平，而且因为在印第安纳州的夫妻家庭通常可以获得公共援助金。但是，根据加里测试的收入支持计划，这些家庭中的许多人有资格获得适度的收入补贴。因此，对加里实验的分析可以深入研究将收入补充计划扩展到工作但低收入家庭的后果。另一方面，无论是丈夫（从事工会工作并拥有资历）还是妻子（不工作或从事使家庭收入高于NIT计划的盈亏平衡水平的工作）都不会被期望显着减少他们的工作努力。

The families with female heads were generally much poorer than the husband-wife families studied. Over 80 percent were receiving welfare benefits from the Aid to Families with Dependent Children (AFDC) program immediately prior to the experiment. About three-fourths of the experimental families that switched from AFDC to the experiment had incomes below the poverty line. The female heads on AFDC at enrollment were very dependent on welfare: 86 percent of their monthly income came
有女性户主的家庭通常比所研究的夫妻家庭贫穷得多。在实验开始前，超过80%的人从“有受抚养子女的家庭援助”（AFDC）计划中获得福利。从AFDC转向实验的实验家庭中，约有四分之三的收入低于贫困线。AFDC的女性校长在入学时非常依赖福利：她们月收入的86%来自福利

7

public transfers, with AFDC grants alone accounting for slightly more than half of their incomes. As with the wives who were studied, only13 percent of the AFDC female heads were employed.
公共转移支付，仅AFDC赠款就占其收入的一半以上。与所研究的妻子一样，只有13%的AFDC女性领导人受雇。

The female-headed families not on AFDC prior to the experiment were somewhat better off--only 38 percent had incomes below the poverty level. Approximately 60 percent of the income of the non-AFDC female-headed families came from earnings (40 percent of the female heads in these families were employed) , while most of the remainder of their income came from Food Stamps, Social Security, and other transfer programs.¹/
在实验之前没有加入AFDC的女性户主家庭的情况要好一些 - 只有38%的收入低于贫困线。在非AFDC女性户主家庭中，大约60%的收入来自收入（这些家庭中40%的女性户主是就业家庭），而其余大部分收入来自食品券、社会保障和其他转移支付计划。

Nevertheless, the female headed families studied were, as a whole,generally quite dependent on public transfers. The few who worked tended to work in low-wage jobs. Thus this group would be expected to be more likely than the husband-wife families to respond to the experiment by reducing their labor supply.
然而，所研究的女户主家庭作为一个整体，总体上相当依赖公共转移支付。少数工作的人往往从事低薪工作。因此，与夫妻家庭相比，预计这一群体更有可能通过减少劳动力供应来对实验做出反应。

The characteristics of the Gary labor market also have important implications for the interpretations and generalizability of the labor supply findings reported here. The Gary labor market is dominated by a few large steel mills plus a number of manufacturing and fabrication plants. This has two important implications. First, in such a highly institutionalized labor market, few opportunities exist for marginal reductions in labor supply, either through reducing hours worked per week or hours worked during the year. Labor supply reductions generally have
加里劳动力市场的特征也对本文报告的劳动力供应调查结果的解释和普遍性具有重要意义。加里劳动力市场由几家大型钢铁厂以及一些制造和制造厂主导。这有两个重要的含义。首先，在这样一个高度制度化的劳动力市场中，无论是通过减少每周工作时间还是减少一年中的工作时间，都几乎没有机会减少劳动力供应。劳动力供应减少通常有

l/A more detailed description of the Gary sam ple can be found in Kehre r (1977).
关于Gary sam ple的更详细描述可以在Kehre r （1977）中找到。

to take the form of withdrawing from employment entirely. The results presented below tend to bear this out. Second, such a labor market contains few jobs traditionally held lby women.¹/ consequently, few females in the experiment worked andthose who did often worked full-time and were more career-oriented (e. g., as teachers) and thus were not the part-time, casual labor which might be expected to respond to the NIT.
采取完全退出工作的形式。下面给出的结果往往证实了这一点。其次，在这样的劳动力市场中，传统上由女性担任的工作很少，因此，实验中很少有女性工作，而那些工作的女性往往全职工作，更注重职业（例如，作为教师），因此不是可能期望对NIT做出反应的兼职临时工。

Clearly, then, the characteristics of the population groups and the local labor market are not identical to those in the U. S. as a whole.Consequently, caution should be exercised before extrapolating the results to a nationwide NIT program. The results we have obtained in analyzing the data suggest, in fact, that these particular population and labor market characteristics did affect the response, as will become clear below.
显然，人口群体和当地劳动力市场的特征与整个美国并不相同。因此，在将结果外推到全国性的NIT计划之前，应谨慎行事。事实上，我们在分析数据时获得的结果表明，这些特定的人口和劳动力市场特征确实影响了反应，这将在下面变得清晰。

1/For example, the Gary SMSA ranked last in an index of labor markets favorable to women devised by Bowen and Finegan (1969, p.774).
1/例如，Gary SMSA在Bowen和Finegan（1969年，第774页）设计的有利于妇女的劳动力市场指数中排名最后。

9

B. AN AVERAGE-TAX-RATE LABOR-SUPPLY MODEL
B. 平均税率劳动力供给模型

A logical first step in estimating the effect of an NIT experiment is to estimate the mean experimental-control group difference:
估计 NIT 实验效果的合乎逻辑的第一步是估计实验组与对照组的平均差异：

L = αT + Xβ + ε, (1)
L = αT + Xβ + ε， （1）

where L is some measure of labor supply, T is an experimental dummy vari-able equal to one if in the experimental group and zero if in the control group, X is a vector of exogenous socioeconomic characteristics, ε is a randomly distributed error term, and a and β are parameters. In such a model the coefficient α measures the "experimental effect", for it represents the mean experimental-control labor-su ppy differential adjusted for differences in X. Estimates of α are presented below.
其中 L 是劳动力供给的某种度量，T 是一个实验虚拟变量，在实验组中等于 1，如果在对照组中等于 0，X 是外生社会经济特征的向量，ε 是随机分布的误差项，a 和 β 是参数。在这样的模型中，系数α衡量“实验效应”，因为它代表了针对X差异调整的平均实验-控制-劳动-supy差异α。

However, estimates of a are only a first step, for the mean experimental-control difference in one particular experiment cannot be generalized to a different population, such as one with a different income distribution and a different percent of families below breakeven. Also,it cannot be generalized to NIT programs with different tax rates and guarantees. For these reasons a more structural model is needed which specifies the particular budget constraint faced by the Gary participants.
然而，对a的估计只是第一步，因为一个特定实验中的平均实验对照差异不能推广到不同的人群，例如收入分配不同和盈亏平衡以下家庭比例不同的人群。此外，它不能推广到具有不同税率和担保的NIT计划。由于这些原因，需要一个更具结构性的模型，该模型指定了Gary参与者面临的特定预算限制。

The standard theory of labor supply suggests that hours of work under an NIT should be a function of the relevant net wage rate and the relevant value of net nonwage income. The NIT benefit is calculated to be equal to [G-t(WH+N)], where G is the quarantee level, t is the tax rate (or "benefit reduction rate"), W is the hourly wage rate, H is hours of work, and N is nonwage income. Since earnings are taxed at the rate t, the net hourly wage rate is W(l-t), and since the NIT benefit
劳动力供给的标准理论认为，NIT下的工作时间应该是相关净工资率和净非工资收入的相关值的函数。NIT福利的计算公式等于[G-t（WH+N）]，其中G是隔离水平，t是税率（或“福利减少率”），W是小时工资率，H是工作时间，N是非工资收入。由于收入按 t 税率征税，因此净小时工资率为 W（l-t），并且由于 NIT 福利

10

at zero hours is (G-tN), total net nonwage income is [G+N(l=t)]. There-fore the labor supply function can be written in implicit form as the following:
零时为（G-tN），非工资净收入总额为[G+N（l=t）]。因此，劳动力供给函数可以用隐式形式写成如下：

H = f[W(l一t) , G+N(l-t) ]
H = f[W（l一t） ， G+N（l-t） ]

Estimation of this function on experimental data can be implemented by setting t and G equal to 0 for controls and by setting t and G equal to the correct values for experiments.
通过将对照组的 t 和 G 设置为 0，将 t 和 G 设置为实验的正确值，可以实现此函数对实验数据的估计。

Unfortunately, the nonlinearity of the NIT budget constraint complicates this estimation. This hours function applies only to individuals with positive benefits and therefore with income less than the breakeven level of income, G/t. For individuals with income greater than G/t,benefits are zero and the marginal NIT tax rate is zero. In a sense, the problem that arises here is "which" tax rate the individual faces. But a more accurate way of describing it is to say that an individual simply faces multiple tax rates--that is, he or she faces a piecewise-linear budget constraint. The problem is compounded for both experimental and control families because they face other tax and transfer programs as well, either during the experiment or before it: the income tax, the payroll tax, a state income tax, Aid to Families with Dependent Children(AFDC), food stamps, and others. Since each of these programs alone contains multiple tax rates, their cumulative effect is a highly nonlinear budget constraint.
不幸的是，NIT预算约束的非线性使这种估计变得复杂。此工时功能仅适用于具有正福利的个人，因此收入低于收入盈亏平衡水平 G/t。对于收入大于G/t的个人，福利为零，边际NIT税率为零。从某种意义上说，这里出现的问题是个人面临的“哪个”税率。但更准确的描述方式是说，一个人只是面临多种税率——也就是说，他或她面临着一个分段线性的预算约束。对于实验家庭和对照家庭来说，这个问题更加复杂，因为他们还面临着其他税收和转移支付计划，无论是在实验期间还是在实验之前：所得税、工资税、州所得税、对有受抚养子女的家庭的援助（AFDC）、食品券等。由于这些计划中的每一个都单独包含多个税率，因此它们的累积效应是一个高度非线性的预算约束。

The most common approach to the problem in the literature has been to estimate some variant of the hours equations H = H[W(1-t') , N'],where W(l-t') is the "local" net wage and N' is the "local" income
在文献中，解决这个问题的最常见方法是估计小时方程 H = H[W（1-t'） ， N']的某种变体，其中 W（l-t'） 是“本地”净工资，N' 是“本地”收入

11

intercept--that is, the net wage and income intercept of the segment upon which the individual is observed (Hall, 1973; Hausman and Wise, 1976;Keeley et al., 1978). Unfortunately, there are two serious problems with this technique. First, the tax rate is clearly endogenous since it is related to hours worked through the tax and benefit formulas. For example,in the case of an NIT, one could obtain a spurious negative tax effect simply because those who work more prior to the experiment (e. g., because of greater tastes for work) are more likely to have a zero local tax rate during the experiment. As a result of this problem, most studies(including those mentioned above) have used some type of instrumental-variables approach.¹/ The second problem with the tech nique is that the function is an incomplete representation of the total labor-supply function, for it only captures the marginal labor-supply choices made within a segment. Implicity it is assumed that an individual first chooses a segment of the piecewise-linear budget constraint, and then chooses a point on the segment. The "choice of segment" and the way the choice is affected by the NIT parameters is not specified.2/
截距——即观察到个人的那部分的净工资和收入截距（Hall，1973;Hausman 和 Wise，1976 年;Keeley等人，1978）。不幸的是，这种技术存在两个严重的问题。首先，税率显然是内生的，因为它与通过税收和福利公式工作的时间有关。例如，在NIT的情况下，人们可能会获得虚假的负税收效应，仅仅是因为那些在实验之前工作更多的人（例如，因为对工作的品味更高）更有可能在实验期间获得零地方税率。由于这个问题，大多数研究（包括上面提到的那些）都使用了某种类型的工具变量方法。隐含性 假设个人首先选择分段线性预算约束的段，然后在段上选择一个点。未指定“段的选择”以及选择受 NIT 参数影响的方式。

To avoid those problems, Burtless and Hausman (1978) have developed an alternative technique which avoids the endogeneity of the tax rate and also estimates the choice-of-segment function. The technique
为了避免这些问题，Burtless和Hausman（1978）开发了一种替代技术，该技术避免了税率的内生性，并估计了段选择函数。技术

¹/For example, Hausman and Wise (1976) followed Rosen (1976) in evaluating all individuals' tax rates at the same, fixed hours point.Keeley et al. (1978) evaluate the tax rate at the hours worked in a previous period (preenrollment).
¹/例如，Hausman 和 Wise （1976） 遵循 Rosen （1976） 评估了所有个人在相同固定时间点的税率。Keeley等人（1978）评估了前一时期（入学前）工作时间的税率。

2/It is also clear that the "choice of segment" and the "choice of hours along a segment" are not separable choices, as the instrumental-variables technique implies. The two should, ideally, be estimated jointly,just as a person jointly chooses them when picking a particular point along the entire budget constraint.
2/同样清楚的是，“段的选择”和“沿段的小时选择”并不是工具变量技术所暗示的可分离的选择。理想情况下，这两者应该一起估计，就像一个人在整个预算约束中选择一个特定点时共同选择它们一样。

12

is based firmly on the correct notion that individuals make utilizy comparisons between segments, but it requires the use of a fairly complex maximum-likelihood procedure. For the purposes of this project,where a very large number of equations of different types must be estimated, a simple r alternative is needed.
它坚定地基于一个正确的概念，即个人在各个部分之间进行利用性比较，但它需要使用相当复杂的最大似然过程。为了这个项目的目的，当必须估计大量不同类型的方程时，需要一个简单的r备选方案。

The method used for our analysis involves approximating the budget constraint rather than representing it in all its detail. Specifically,we smooth the budget constraint by averaging all the marginal tax rates along it. The cumulative tax rate in each segment of the individual's constraint is weighted by its fraction of hours worked covered, and then summed to obtain a weighted-average tax rate. This approach has the advantage of modeling the response to the entire budget constraint, not just to a local segment of it (although it only approximates the entire constraint) and of using a tax rate that is not endogenous. However,since it only approximates the (entire) budget constraint, it must a fortiori be considered only an approximation to the true labor-supply function.¹/
我们用于分析的方法涉及近似预算约束，而不是详细表示它。具体来说，我们通过平均沿线的所有边际税率来平滑预算约束。个人约束的每个部分的累积税率由其所涵盖的工作小时数加权，然后相加得到加权平均税率。这种方法的优点是模拟了对整个预算约束的响应，而不仅仅是对预算约束的局部部分（尽管它只近似于整个约束），并且使用非内生的税率。然而，由于它只是近似于（整个）预算约束，因此它只能被视为对真实劳动力供给函数的近似。

The implications of the method can be seen by specifying the labor supply function linearly:
通过线性指定劳动力供给函数，可以看出该方法的含义：

H = γ + δW(l - r- t) + η(N+B。), (2)
H = γ + δW（l - r- t） + η（N+B。), (2)

Where "r" is the average tax rate on non-NIT income and "t" is the average tax rate on NIT income. NIT benefits at zero hours are BO, equal to(G-tN). If. an individual has a sufficiently high wage rate that he or
其中“r”是非NIT收入的平均税率，“t”是NIT收入的平均税率。零小时时的 NIT 收益为 BO，等于 （G-tN）。如果。一个人的工资率足够高，他或

l/see Moffitt and Kehrer (forthecoming) for a full discussion of all the models used in the experiments to address the problem.
l/参见 Moffitt 和 Kehrer （forthecoming） 对实验中用于解决该问题的所有模型的全面讨论。

13

she is above breakeven over some portion of the contraint, the averaging procedure assigns to such an individual a value of "t" which is a weighted average of the positive, below-breakeven NIT tax rate and the zero, above-breakeven tax rate. For experimentals who are below breakeven over the entire hours range, the average NIT tax rate "t" is the same as the below-breakeven tax rate. Thus, unlike some other experimental models, an individual who may "above breakeven" aat preenrollment (i. e., located on the upper portion of the constraint) is allowed to have a response, although it is assumed that the stimulus (i. e., the average tax rate) is lower than for someone who is below breakeven at all hours points.
她在禁忌的某些部分高于盈亏平衡，平均程序为这样的人分配了一个值“t”，该值是正的、低于盈亏平衡的 NIT 税率和零、高于盈亏平衡的税率的加权平均值。对于在整个小时范围内低于盈亏平衡的实验者，平均NIT税率“t”与低于盈亏平衡的税率相同。因此，与其他一些实验模型不同，一个可能在预注册时“高于盈亏平衡”（即位于约束的上部）的人被允许有反应，尽管假设刺激（即平均税率）低于在所有时间点低于盈亏平衡点的人。

The same averaging procedure is followed for non-NIT tax rates.Federal income taxes, for example, impose increasing marginal taxes over the budget line in a series of segments corresponding to tax brackets.AFDC benefits, like those for an NIT, fall to zero at some point and create a nonlinearity in the constraint. The average tax rate "r" is calculated for each individual by constructing the individual's entire set of marginal tax rates at all hours points and by then calculating their average.
非 NIT 税率遵循相同的平均程序。例如，联邦所得税在与税级相对应的一系列部分中，在预算项目上征收增加的边际税。AFDC 收益（如 NIT 的收益）在某些时候降至零，并在约束中产生非线性。每个人的平均税率“r”是通过构建个人在所有时间点的整套边际税率，然后计算其平均值来计算的。

The extent to which this approximation of the contraint yields biased coefficients is an empirical question. Averaging the tax rates over the contraint throws away much information on marginal tax rates along the constraint, but the importance of these tax rates is partially affected by the opportunity to adjust hours marginally. On a priori grounds, one would expect that "intramarginal" tax rates would be less important for individuals who are working in highly structured labor markets which make marginal adjustments in hours of work difficult. As has been
这种近似值在多大程度上产生偏置系数是一个经验问题。对约束值的税率进行平均会丢弃许多关于边际税率的信息，但这些税率的重要性在一定程度上受到边际调整时间的机会的影响。在先验的基础上，人们会认为“边际内”税率对于在高度结构化的劳动力市场中工作的个人来说不那么重要，这使得工作时间的边际调整变得困难。一如既往

14

mentioned before, the Gary labor market is indeed highly structured and offers few part-time jobs to its population; thus it is not implausible that many individuals respond only to the average tax rate on earnings.This institutional feature of the local labor market provides, in fact,the major motivation for our development of an average-tax-rate model.
前面提到，加里劳动力市场确实是高度结构化的，为其人口提供的兼职工作很少;因此，许多人只对收入的平均税率做出反应并非难以置信。事实上，当地劳动力市场的这一制度特征为我们开发平均税率模型提供了主要动力。

Equation (2) is modified slightly, but an important way, to allow NIT and non-NIT variables to have different coefficients:
等式 （2） 略有修改，但是一种重要方法，允许 NIT 和非 NIT 变量具有不同的系数：

H =γ + δ'W(l-r) + δ"(-Wt) + η'N + n"Bo- (3)
H =γ + δ'W（l-r） + δ“（-Wt） + η'N + n”Bo- （3）

This separation is important because it avoids "contaminating" the effect of the experimental stimulus with the effect of nonexperimental stimuli. If equation (2) were estimated, one could easily obtain estimates of δ and η which are, in large part, a result of behavioral responses to nonexperimental variables. This would defeat one of the main purposes of running an experiment, which is to obtain estimates different than those in nonexperimental data. Of course the significance of the coefficient differences is an empirical question, which we will examine below. However, despite the potential importance of separating the coefficients, this is the first experimental study to do so in a formal way.
这种分离很重要，因为它避免了用非实验刺激的影响“污染”实验刺激的效果。如果估计方程（2），可以很容易地获得δ和η的估计，这在很大程度上是对非实验变量的行为反应的结果。这将破坏运行实验的主要目的之一，即获得与非实验数据不同的估计值。当然，系数差异的意义是一个经验问题，我们将在下面进行研究。然而，尽管分离系数具有潜在重要性，但这是第一个以正式方式这样做的实验研究。

Given our particular resolution of the kinked-budget-line problem,there nevertheless remain a number of important econometric and specification issues that must be addressed. The actual empirical implementation of equation (3) on the particular data set we have available is discussed in the next section.
鉴于我们对扭结预算线问题的特殊解决方案，仍然存在一些重要的计量经济学和规格问题必须解决。下一节将讨论公式（3）在现有特定数据集上的实际实证实现。

15

C. EMPIRICAL-SPECIFICATION ISSUES
C. 经验规格问题

The data base available for the estimation of equation (3) is drawn from the interview s that were conducted periodically throughout the experiment. One preenrollment interview and seven during-experiment interview s provide the basic panel of information on income, labor supply,and demographic characteristics. Each variable is measured in monthly terms as of the month of interview. The selection of an analysis sample from the full set of enrolled individuals is discussed further below.
可用于估计方程式（3）的数据库来自在整个实验过程中定期进行的访谈。1次入学前访谈和7次实验期访谈提供了关于收入、劳动力供给和人口特征的基本信息面板。每个变量都是从访谈当月开始按月衡量的。从全部入组个体中选择分析样本的问题将在下面进一步讨论。

To estimate equation (3), the specification and econometric issues that must be resolved include the definition of the dependent variable, the choice of sample, the choice of estimating technique, and other issues. As several previous studies have shown, coefficient estimates may be quite sensitive to some of these decisions (e. g., Cain and Watts, 1973; DeVanzo et al., 1976). Unfortunately, while there are well-defined advantages and disadvantages to many alternative specifi-cations and procedures, their net quantitative impacts are usually unknown and consequently there is no single "correct" way to proceed.Therefore, in this study we have chosen to use a variety of specifi-cations and to conduct a large number of sensitivity tests, the nature of which will become clearer below. We will discuss each specification issue in turn and will (1) choose a preferred specification and (2) define a set of reasonable alternatives that we also estimate. The equation that is specified in the preferred way in all respects is called the"basic" model, and will be the benchmark equation around which all the sensitivity tests will revolve.
要估计方程（3），必须解决的规格和计量经济学问题包括因变量的定义、样本的选择、估计技术的选择以及其他问题。正如先前的几项研究表明的那样，系数估计可能对其中一些决策非常敏感（例如，Cain and Watts，1973;DeVanzo 等人，1976 年）。遗憾的是，虽然许多替代性规范和程序都有明确的优点和缺点，但它们的净数量影响通常是未知的，因此没有单一的“正确”方法进行。因此，在这项研究中，我们选择使用各种规格并进行大量的灵敏度测试，其性质将在下文中变得更加清晰。我们将依次讨论每个规格问题，并将 （1） 选择首选规格和 （2） 定义一组我们也估计的合理替代方案。在所有方面都以首选方式指定的方程称为“基本”模型，并且将是所有灵敏度测试将围绕的基准方程。

16

n

1

l. Dependent Variable
l. 因变量

The dependent variable used in the basic model is hours of work per month, zeros included. As a sensitivity test an equation is estimated with an employment-status dependent variable--equal to one if employed and zero if not--as an indirect test of whether hours of work are flexible enough to allow individuals to respond marginally to the experiment. If hours are completely inflexible and individuals can respond only by lowering their probability of being employed, the coefficients in the hours equation should be equal to H times the coefficients in the employment-status equation, where H is the mean hours worked of workers. As has been mentioned before, the nature of the Gary labor market may generate such a result.
基本模型中使用的因变量是每月的工作小时数，包括零。作为敏感性检验，使用就业状况因变量（如果就业则等于 1，如果不就业则为 0）估计方程，作为间接检验工作时间是否足够灵活，允许个体对实验做出轻微反应。如果工作时间完全没有灵活性，个人只能通过降低受雇的可能性来做出反应，那么工时方程中的系数应等于H乘以就业状况方程中的系数，其中H是工人的平均工作时间。如前所述，加里劳动力市场的性质可能会产生这样的结果。

As a second sensitivity test, an equation is estimated with a"validated" employment-status dependent variable to test whether any underreporting of employment in the interview s affects the estimated experimental response. Presumably any differential underreporting by experimentals and controls would cause an overly-large labor-supply effect, since experimentals have an incentive to underreport hours in order to increase NIT payments. The validated employment-status variable is created by using data on individual earnings collected from the Indiana Department of Employment Security, to whom employers send earnings reports for certification of eligibility for unemployment insurance. These data may be more accurate than the self-reported data collected in the experimental interview s because it is reported by the employer rather than the individual. Note, however, that this gives us an outside measure only of employment status; hours of work are not
作为第二个敏感性检验，使用“已验证”的就业状况因变量估计方程，以检验访谈中任何就业报告不足是否会影响估计的实验响应。据推测，实验者和对照组的任何差异性少报都会导致过大的劳动力供给效应，因为实验者有动机少报工时以增加NIT支付。经过验证的就业状态变量是使用从印第安纳州就业保障部收集的个人收入数据创建的，雇主向该部门发送收入报告以证明是否有资格获得失业保险。这些数据可能比实验访谈中收集的自我报告数据更准确，因为它是由雇主而不是个人报告的。然而，请注意，这仅为我们提供了就业状况的外部衡量标准;工作时间不是

17

reported by employers. In addition, since many low-income workers are not covered by unemployment insurance, their earnings are not reported to the Department. A separate paper by Greenberg et al. (1979)addresses this problem of coverage in more detail; it is ignored here.
由雇主报告。此外，由于许多低收入工人没有参加失业保险，他们的收入没有向该部报告。Greenberg等人（1979）的另一篇论文更详细地解决了这个覆盖率问题;这里忽略了它。

2. Budget Line
2. 预算项目

The construction of the budget line and the average tax . rate uses an individual's gross hourly wage rate and amount of non-work-conditioned nonwage income together with the formulas of the tax and transfer programs for which he or she is eligible. The formulas for two tax programs, the federal income tax and the social security payroll tax, and for two transfer programs, AFDC and the NIT, are involved.¹/ Both the NIT andAFDC reimburse positive. taxes, and this must be taken into account in the formulas. In addition, the NIT taxes some forms of income at a 100percent rate instead of at a rate of t. The details of the computations are shown in Appendix B.
预算线和平均税额的建设。税率使用个人的小时总工资率和非工作条件的非工资收入金额，以及他或她有资格参加的税收和转移支付计划的公式。涉及两个税收计划（联邦所得税和社会保障工资税）以及两个转移计划（AFDC 和 NIT）的公式。税收，这在公式中必须考虑到。此外，NIT 以 100% 的税率而不是 t 的税率对某些形式的收入征税。计算的详情见附录 B。

The computation of the average tax rates is very simple once the end points of the budget constraint are chosen, for the average net wage rate is simply the slope of the line drawn between the endpoints.For the Gary sample, the endpoints chosen are zero hours of work and 173hours of work per month, the latter being approximately equal to full-time work. This choice is based upon the tabulated characteristics of the
一旦选择了预算约束的终点，平均税率的计算就非常简单，因为平均净工资率只是终点之间划定的线的斜率。对于 Gary 样本，选择的端点是每月零工作小时和 173 小时工作，后者大约等于全职工作。这种选择是基于

¹/The Indiana income tax is ignored because it is very small in magnitude. Food stamps are also ignored in the basic model because they are in-kind income and thus do not play the same role as cash income in the theory. However, as discussed below, food stamps are included as a sensitivity test.
¹/印第安纳州所得税被忽略，因为它的规模非常小。食品券在基本模型中也被忽略了，因为它们是实物收入，因此在理论中起不到与现金收入相同的作用。但是，如下所述，食品券被列为敏感性测试。

18

□

∫

卩

『

乁

!

①

1

し

This decision is based upon previous research by Wolf (1977) which showed that the Gary experiment had no effect on marital dissolution,implying that marital status can be treated as exogenous to the experiment.Third, families who left the experiment ("attriters") are included in the sample for periods before their departure. This decision is based upon research by Hausman and Wise (1979) which indicates that there is little attrition bias in the Gary sam ple in a structural supply equation(i. e., one which includes the usual right-hand-side taste variables-these control for attrition effects). Nevertheless, despite our a priori notions, all three of these sam ple selection decisions are subjected to sensitivity testing by reestimating the equation by different criteria (see below).
这一决定是基于Wolf（1977）先前的研究，该研究表明Gary实验对婚姻解除没有影响，这意味着婚姻状况可以被视为实验的外生因素。第三，离开实验的家庭（“Attriters”）在他们离开前的一段时间内被包括在样本中。这一决定是基于Hausman和Wise（1979）的研究，该研究表明，在结构供应方程中，Gary sam ple几乎没有损耗偏差（即，包括通常的右侧味觉变量 - 这些对损耗效应的控制）。然而，尽管我们有先验的概念，但所有这三个 sam ple 选择决策都通过通过不同的标准重新估计方程来进行敏感性测试（见下文）。

5. Time Periods Examined
5. 审查的时间段

For the basic equation all seven during-experiment interviews are included in the data set rather than selecting only those in the middle year or middle two years of the experiment. In many analyses of other experiments' data, only the middle periods of the data have been used because it is suspected that this most nearly captures the long-run response to an NIT. Early periods may be affected by adjustment and start-up factors, while late periods may be affected by readjustments in anticipation of the end of the experiment. In theGary data, however, there appears to be no such pattern in the mean experimental-control differences. The pattern of differences, reported in Appendix E, shows very little variation over time in the response and sometimes even shows a greater response in the earlier or later periods of the experiment than in the middle. Nevertheless, since these
对于基本方程式，所有七个实验期间的访谈都包含在数据集中，而不是只选择实验中年或中间两年的访谈。在对其他实验数据的许多分析中，只使用了数据的中间周期，因为怀疑这最接近于捕获对NIT的长期响应。早期可能会受到调整和启动因素的影响，而后期可能会受到预期实验结束的重新调整的影响。然而，在Gary数据中，平均实验对照差异似乎没有这种模式。附录 E 中报告的差异模式显示，响应随时间的变化非常小，有时甚至在实验的早期或后期比中间阶段表现出更大的响应。然而，由于这些

20

mean experimental-control-differences may not hold up in a structural model, we have also estimated our equation on the middle period of the experiments. as a sensitivity test.
在结构模型中，平均实验-对照-差异可能不成立，我们还估计了实验中间期的方程。作为敏感性测试。

Another sensitivity test to our use of seven monthly data points is afforded by the use of quarterly data, which were made available late in the analysis. These data, formed by averaging a continuous monthly data file into quarters, represent more observations and a more continuous profile. The efficiency of the estimates, if not their consistency,should be improved by reestimation on this data set.
我们使用七个月度数据点的另一个敏感性测试是通过使用季度数据提供的，这些数据在分析后期可用。这些数据是通过将连续的月度数据文件平均为季度而形成的，代表了更多的观测值和更连续的轮廓。估计的效率，如果不是它们的一致性，应该通过重新估计这个数据集来提高。

6. Sample Stratification and Preenrollment Differences
6. 样本分层和入组前差异

One of the problems in using the data from the Gary experiment is that the sample was stratified by income level and that the experimental randomization took place only within these income strata(Conlis k and Watts, 1969). Between strata, different experimental-control allocations were chosen according to a criterion function developed by Conlisk and Watts. The result of this stratification,which was followed in the other experiments as well, is that there are occasionally experimental-control differences in labor supply prior to the experiment.
使用Gary实验数据的问题之一是样本按收入水平分层，并且实验随机化仅在这些收入阶层内进行（Conlis k和Watts，1969）。在各层之间，根据 Conlisk 和 Watts 开发的标准函数选择不同的实验对照分配。这种分层的结果，也适用于其他实验，即在实验之前，劳动力供应偶尔存在实验对照差异。

The work by Hausman and Wise (1977) in this area provides evidence that the stratification had little impact in the Gary experiment.Hausman and Wise showed that the different sampling ratios by income level can be modeled as a straightforward type of selection bias and thatTobit-like maximum-likelihood procedures can correct it. In their empirical work on the Gary data, however, the actual empirical magnitude
Hausman和Wise（1977）在这一领域的工作提供了证据，证明分层在Gary实验中影响不大。Hausman和Wise表明，按收入水平划分的不同抽样比率可以建模为一种直接的选择偏差，而类似Tobit的最大似然程序可以纠正它。然而，在他们对加里数据的实证研究中，实际的经验量级

21

of the difference between correctly-estimated coefficients and biased,OLS coefficients is quite small. As iit turns out, this may be because(ironically) the stratification was implemented very poorly in Gary, for very inaccurate measures of income were used in the computations.
在正确估计的系数和有偏差的系数之间的差异中，OLS系数相当小。事实证明，这可能是因为（具有讽刺意味的是）加里的分层实施得非常糟糕，因为在计算中使用了非常不准确的收入衡量标准。

Nevertheless, experimental-control differences did occur in some of our preenrollment-estimated equations. This may be a result of some residual stratification effect or it may be a result of other factors-either from the partial failure of the randomization or from sample deletions that cause nonrandom experimental-control differences. In any case, we therefore control for possible preenrollment differences in all our equations. There are basically two different ways of doing so. The first is to estimate both preenrollment and during-experiment experimental-control labor-supply differences, and to measure the effect of the experiment as the change in this difference from the earlier to the later period. This procedure has a familial resemblance to the first-difference method. The variation of it which we use is to pool both preenrollment and during-experiment observations into a single equation and to "net out"preenrollment differences. For example, in the context of the treatment-difference model considered in equation (1) some while back, this "netting out" procedure involves estimating the equation:
尽管如此，在我们的一些入组前估计方程中确实发生了实验对照差异。这可能是一些残余分层效应的结果，也可能是其他因素的结果 - 要么是随机化的部分失败，要么是导致非随机实验对照差异的样本缺失。因此，在任何情况下，我们都控制了所有方程式中可能存在的入学前差异。基本上有两种不同的方法可以做到这一点。首先是估计入学前和实验期间的实验控制劳动力供给差异，并衡量实验的影响，作为这种差异从早期到后期的变化。该过程在家族上与第一差值方法相似。我们使用的它的变体是将入组前和实验期间的观察结果汇集到一个方程中，并“剔除”入组前的差异。例如，在前段时间等式（1）中考虑的处理-差异模型的背景下，这种“净额计算”过程涉及估计等式：

L = αT + βTD + δD + ε,
L = αT + βTD + δD + ε，

where T is the previously-defined treatment dummy and D is a dummy equal to l if the observation on L is during the experiment and o if it is prior to the experiment. The coefficient β measures the experimental effect (i. e., the experimental-control difference net of preenrollment differences, as measured by α ).
其中 T 是先前定义的处理假人，D 是等于 l 的假人，如果对 L 的观察是在实验期间，则 o 是在实验之前。系数β衡量实验效果（即，通过α测量的入组前差异减去实验对照差异）。

22

The second method is to estimate a during-experiment equation including the preenrollment value of the dependent variable as an independent variable. For convenience of exposition, the first method is labeled the "Net" model and the latter is labeled the "Lagged" model.As a rough rule of thumb, the Net model is relatively more desirable when the preenrollment selection criterion is related to unobserved variables,and the Lagged model is preferred when the criterion is related to observed variables (see Appendix C).In the basic model of this paper,the Net model is used because (1) the observed variable usually presumed to cause preenrollment differences in the NIT experiments is the income variable used in the sample income stratification but, as mentioned before, the stratification had little effect in Gary; and (2) the lagged dependent variable may bias the coefficient on the wage rate, which happens to be interacted with a treatment variable (the tax rate).Nevertheless, as a sensitivity test,the Lagged model is also estimated.
第二种方法是估计实验期间方程，包括因变量的预注册值作为自变量。为了便于说明，第一种方法被标记为“Net”模型，后者被标记为“滞后”模型。粗略的经验法则是，当入组前选择标准与未观察到的变量相关时，Net 模型相对更可取，而当标准与观察到的变量相关时，滞后模型更受欢迎（见附录 C）。在本文的基本模型中，使用Net模型是因为（1）在NIT实验中通常被认为会导致入组前差异的观察变量是样本收入分层中使用的收入变量，但如前所述，分层对Gary的影响很小;（2）滞后因变量可能会使系数偏向工资率，而工资率恰好与处理变量（税率）相互作用。然而，作为灵敏度测试，滞后模型也被估计。

7. Wage Rates
7. 工资率

Missing data for wage rates of nonworkers is a standard problem in labor-supply studies, and can be partly solved by using an instrument for the wage rate or by adopting more complex maximum-likelihood procedures(Heckman, 1974). In an experiment there is the additional problem that the NIT may affect individual wage rates; therefore, the experimental response may be missed if the during-experiment wage rate is used as an independent variable. In the basic model used below, actual preenrollment wage rates are used for workers and a predicted preenrollment wage rate is used for nonworkers. This asymmetry is allowed in order to increase
非工人工资率的缺失数据是劳动力供给研究中的一个标准问题，可以通过使用工资率工具或采用更复杂的最大似然程序来部分解决（Heckman，1974）。在实验中，还有一个额外的问题，即NIT可能会影响个人工资率;因此，如果将实验期间的工资率用作自变量，则可能会错过实验响应。在下面使用的基本模型中，实际的注册前工资率用于工人，预测的注册前工资率用于非工人。这种不对称性是允许的，以便增加

23

the variance in the wage-rate variable, for predicted wage rates generally have relatively little variance. On the other hand, some bias may result from this procedure, for it introduces an error term into the wage-rate variable which is related to the dependent variable (i. e.,whether working or not). Therefore, as a sensitivity test, the equation is also estimated with the predicted preenrollment wage used for all workers. In addition, a sensitivity test to the use of preenrollment values in general is conducted by estimating the equation using contem-poraneous wage rates. If the experiment has an effect on wages, the coefficient estimates may differ. See Appendix D for the detailed wage-predicting equations.
工资率变量的方差，对于预测的工资率，通常具有相对较小的方差。另一方面，这个过程可能会导致一些偏差，因为它在工资率变量中引入了一个误差项，该误差项与因变量（即，是否工作）有关。因此，作为敏感性检验，该方程也是使用用于所有工人的预测预注册工资来估计的。此外，通过使用连续工资率估计方程，对一般使用预注册值的敏感性测试。如果实验对工资有影响，则系数估计值可能会有所不同。有关详细的工资预测公式，请参阅附录 D。

8. Interdependence of Family Labor-Supply Decisions
8. 家庭劳动力供给决策的相互依存关系

The model developed in the previous sections is a model of individual labor supply and has not included the possibility that the individual may be a member of a household which contains more than one potential worker. However, the labor-supply decisions of household members may be interdependent. Family labor supply decisions have been analyzed in the context of an NIT by Killingsworth (1976). In the most general non-tax case, the wage rates of all employable household members should appear as exogenous variables in the labor supply equation of each employable person. If contemporaneous earnings of other household members are entered instead, two types of bias may arise. First, the presence of nonzero cross-substitution effects will bias the coefficient.However, as Killingsworth notes, the available empirical studies do not agree on the sign and magnitude of the cross-substitution effect. Second,
前几节开发的模型是个人劳动力供给模型，不包括个人可能是包含多个潜在工人的家庭成员的可能性。然而，家庭成员的劳动力供应决策可能是相互依存的。Killingsworth（1976）在NIT的背景下分析了家庭劳动力供应决策。在最一般的非税收情况下，所有可就业家庭成员的工资率应作为外生变量出现在每个就业人口的劳动力供给方程式中。如果将其他家庭成员的同期收入计算在内，则可能会出现两种类型的偏见。首先，非零交叉替换效应的存在会使系数产生偏差。然而，正如基林斯沃思所指出的，现有的实证研究并未就交叉替代效应的符号和大小达成一致。第二

24

even if cross-substitution effects are zero, a second type of bias may arise from the simultaneity that works through the income effect alone.
即使交叉替代效应为零，第二种类型的偏差也可能来自仅通过收入效应起作用的同时性。

The problem is greatly complicated when tax and transfer programs such as the NIT are introduced. In the general tax case, the n tax rates in an individual's budget constraint (t₁' t₂' ……… , tn) are a function of other family members' labor supplies; thus the net wages are no longer exogenous. Even the drastically simplified, linearized budget constraint we use here is subject to the same problem, for the tax rates over the zero-to-full-time range which are averaged for an individual are different according to the value of nonwage income and the income of the spouse.
当引入税收和转移支付计划（例如NIT）时，问题就变得非常复杂。在一般税收案例中，个人预算约束中的n税率（t₁'t₂'.........，tn）是其他家庭成员的劳动供给的函数;因此，净工资不再是外生的。即使是我们在这里使用的大幅简化、线性化的预算约束也会遇到同样的问题，因为根据非工资收入的价值和配偶的收入，在零到全职范围内的平均税率是不同的。

Our approach is to estimate a structural rather than reduced-form equation and to take into account only the interdependence of labor supply of the spouse. We use an instrumental variable for the earnings of the spouse, equal to the spouse's preenrollment wage rate times a fixed number of hours worked for the entire sample, equal to the mean. Thus in the husbands' equations, we add to his nonwage income the value of his wife's wage rate times 95 hours per month, the mean in the wives'sample. In the wives' equations, we add to her nonwage income a comparable instrumental variable for the husbands' earnings. In all equations, including those for female heads, the actual, contemporaneous earnings of "tertiary" family members is added to nonwage income.
我们的方法是估计一个结构方程而不是简化形式的方程，并且只考虑配偶劳动力供给的相互依赖性。我们使用一个工具变量来表示配偶的收入，等于配偶的入学前工资率乘以整个样本的固定工作小时数，等于平均值。因此，在丈夫的方程式中，我们将他妻子的工资率乘以每月 95 小时的值加到他的非工资收入中，即妻子样本中的平均值。在妻子的方程式中，我们在妻子的非工资收入中增加了一个与丈夫收入相当的工具变量。在所有方程式中，包括女性户主的方程式中，“第三级”家庭成员的实际同期收入都加在非工资收入中。

For wives we also test a model with the "male chauvinist"assumption (Killingsworth, 1976) that wives take their husbands' earnings as exogenous, although not vice-versa. Husbands' actual, contemporaneous earnings are added to the wife's nonwage income. The results from this model and the one above may bracket the true effects for wives.
对于妻子，我们还测试了一个具有“男性沙文主义”假设（Killingsworth，1976）的模型，即妻子将丈夫的收入视为外生收入，尽管反之亦然。丈夫的实际同期收入被添加到妻子的非工资收入中。这个模型和上面的模型的结果可以括起来对妻子的真正影响。

25

9. Other Independent Variables
9. 其他自变量

Beside the four budget-constraint variables of equation (3),independent variables are included for the number of adults in the family,the total number of children, and the presence of children in various age ranges as dummy variables, which all together proxy several different effects (the need for home time, the availability of other people to supervise children, the number of people needing financial support, etc.).The local SMSA unemployment rate and the season (one if summer, zero if not) at the time of the interview are included to control for cyclical and seasonal variations in employment in the Gary SMSA. A variable for the presence of a multiple-family household--i. e., the presence of a separate family in the household which may share income, rent, expenses,etc.--is included because such families may be receiving some financial support not captured by the income variables in the equation. A number of these households were enrolled in the experiment.
除了等式（3）的四个预算约束变量外，还包括家庭中成年人的数量、儿童总数和不同年龄段的儿童的存在等自变量作为虚拟变量，这些变量共同代表了几种不同的影响（需要家庭时间、是否有其他人来监督孩子、 需要经济支持的人数等）。包括面试时当地的 SMSA 失业率和季节（如果夏季为 1，则为零），以控制 Gary SMSA 就业的周期性和季节性变化。多户家庭存在的变量 - i。例如，家庭中是否存在一个可能分担收入、租金、支出等的单独家庭，因为这些家庭可能正在接受等式中收入变量未涵盖的一些经济支持。这些家庭中的一些家庭参加了实验。

10. other Issues
10. 其他问题

Limited Duration of the Experiment. As many observers have noted,the limited-duration problem may be the greatest disadvantage of the income maintenance experiments. Some commentators have automatically assumed, however, that the response in the experiment will be therefore smaller than that in a permanent national program, but there are actually effects that work the other way (Metcalf, 1973; Ashenfelter, 1978). If income effects are negative (i. e., leisure is a normal good), it is true that a permanent program would provide many more years of benefits and thus would generate larger labor-supply reductions. But if intertemporal substitution effects are nonzero, individuals will also tend to take
实验的持续时间有限。正如许多观察家所指出的那样，有限期限问题可能是收入维持实验的最大缺点。然而，一些评论家自动假设，因此，实验中的反应将小于永久性国家计划中的反应，但实际上存在以相反方式起作用的效应（Metcalf，1973;Ashenfelter，1978 年）。如果收入效应是负面的（即，闲暇是一种正常的商品），那么永久性计划确实会提供更多年的福利，从而产生更大的劳动力供应减少。但是，如果跨期替代效应不为零，个体也会倾向于采取

26

advantage of the short duration of the reduced price of leisure (relative to a permanent program) and will tend to overrespond. The net effects of these two opposing forces is ambiguous.
休闲价格降低的短时间优势（相对于永久性计划），并且往往会反应过度。这两种对立力量的净效应是模棱两可的。

Also, as Rees and Watts (1975) point out, the net effect of a permanent NIT may be understated by a short duration experiment if it is costly to find a job after a worker quits or if it is difficult to change hours marginally. For example, adult males have a strong attachment to the labor force and may not wish to undertake the risk involved in quitting a job during a short-duration NIT.
此外，正如Rees和Watts（1975）所指出的那样，如果工人辞职后找工作的成本很高，或者如果很难边际改变工作时间，那么永久性NIT的净效应可能会被短期实验低估。例如，成年男性对劳动力有强烈的依恋，可能不希望承担在短期 NIT 期间辞职所涉及的风险。

However, we believe that the biases introduced by limited duration are probably relatively minor. For example, Metcalf (1974) developed a method of using consumption data to estimate the pure life-cycle bias under costless adjustment, and found that the biases were small--2 to 6percent in the substitution (tax) effect and 8 to 27 percent in the income(guarantee) effect. Perhaps more accurate estimates are those from theSeattle-Denver experiment, where both three-year and five-year families were enrolled. Some results (Burtless and Greenberg, 1978; Keeley et al.,1978; Robins and West, 1978) indicate a substantial but insignificant difference in response between the groups. The most formal model (Moffitt,1979) indicates that there is a slight underresponse in the 3-year sample.As for costs of adjustment, it appears that they are significant (Robins and West, 1978). However, a three-year experiment does not appear to seriously understate the response, for most of the labor-supply adjustment takes place within a two-to-four-year period.
然而，我们认为，有限期限带来的偏见可能相对较小。例如，Metcalf（1974）提出了一种利用消费数据来估计无成本调整下纯生命周期偏差的方法，发现偏差很小——替代（税收）效应为2%至6%，收入（保证）效应为8%至27%。也许更准确的估计是来自西雅图 - 丹佛实验的估计，该实验招募了三年和五年制家庭。一些结果（Burtless和Greenberg，1978;Keeley等人，1978;Robins和West，1978）表明两组之间的反应存在显着但微不足道的差异。最正式的模型（Moffitt，1979）表明，在3年的样本中存在轻微的响应不足。至于调整成本，它们似乎是巨大的（Robins和West，1978）。然而，为期三年的实验似乎并没有严重低估这种反应，因为大多数劳动力供给调整都是在两到四年内进行的。

27

Hawthorne Effect. A second problem is the possibility of aHawthorne effect which occurs if the experimental subjects react to the act of being studied itself rather than to the experimental treatment,perhaps because of repeated interviewing or from publicity in the local media. In the usual case of such effects, the subjects learn the objective of the study and possibly the biases of the investigators from the inter-viewing and/or the publicity, and react to this knowledge itself, perhaps by changing their behavior so as to fulfill the prior expectations of the investigators.
霍桑效应。第二个问题是霍桑效应的可能性，如果实验对象对被研究的行为本身而不是对实验治疗做出反应，就会发生霍桑效应，这可能是因为重复采访或当地媒体的宣传。在这种影响的通常情况下，受试者从相互观看和/或宣传中了解研究的目的以及可能的研究者的偏见，并对这种知识本身做出反应，也许通过改变他们的行为来满足研究者的先前期望。

Hawthorne effects are not likely to be a serious problem in the income maintenance experiments for two reasons. First, to the extent that such effects are additive and affect both experimentals and controls(as repeated interviewing and publicity do), the experimental-control difference may be unaffected. Second, Hawthorne effects are usually only found in small-scale social experiments where there is extensive direct contact between the investigators and the participants. In a large urban experiment with over a thousand subjects, in which there is little contact between subjects and investigators, and where the benefit-determination process is relatively impersonal (families mail in an income report form and receive a check), it is implausible that the subjects would internalize the perceived norms of the investigators and adjust their labor supply accordingly.
霍桑效应在收入维持实验中不太可能成为一个严重的问题，原因有二。首先，在某种程度上，这种效应是累加的，并且同时影响实验和对照组（如重复访谈和宣传），实验组与对照组的差异可能不受影响。其次，霍桑效应通常只存在于小规模的社会实验中，在这些实验中，研究人员和参与者之间有广泛的直接接触。在一项涉及一千多名受试者的大型城市实验中，受试者和研究者之间几乎没有接触，而且利益确定过程相对非个人化（家庭邮寄收入报告表并收到支票），受试者不太可能内化研究者的感知规范并相应地调整他们的劳动力供应。

Income Truncation. Finally, a potential problem in all the experiments was the income truncation of the sample. In the New Jersey experiment, for example, no families with income greater than 1.5 times
收入截断。最后，所有实验中的一个潜在问题是样本的收入截断。例如，在新泽西州的实验中，没有收入超过1.5倍的家庭

28

the poverty line were enrolled, causing substantial biases in the response (Hausman and Wise, 1976). When this issue was examined in the Gary experiment, however, the bias was found to be small and empirically unimportant (Hausman and Wise, 1977). The reason for this appears to be that a relatively large number of high income families were enrolled. Although smaller in number than low-income families,they were apparently great enough to fill out the income distribution and eliminate any truncation bias.
贫困线被纳入，导致反应出现重大偏差（Hausman和Wise，1976）。然而，当在Gary实验中检查这个问题时，发现偏差很小，而且在经验上并不重要（Hausman和Wise，1977）。其原因似乎是招收了相对多的高收入家庭。虽然数量少于低收入家庭，但它们显然足以填补收入分配并消除任何截断偏差。

D. BASIC MODEL RESTATED
D. 基本模式重述

Having discussed the empirical specification at some length it will be useful to restate what we have called the "basic" model. The equation to be estimated is the following:
在详细讨论了经验规范之后，重申我们所说的“基本”模型将是有用的。要估计的等式如下：

(4)
（四）

where
哪里

H = hours of work per month
H = 每月工作小时数

w = hourly wage rate
W = 小时工资率

t = average NIT tax rate
t = 平均 NIT 税率

D = during-experiment dummy variable, equal to l if a during-experiment observation and 0 otherwise
D = 实验期间虚拟变量，如果是实验期间观察值，则等于 l，否则为 0

Bo= NIT benefit at zero hours
Bo= 零小时时的 NIT 收益

r = average non-NIT tax rate
r = 平均非 NIT 税率

N = non-NIT income at zero hours
N = 零小时时的非 NIT 收入

X = vector of other variables.
X = 其他变量的向量。

29

As explained above, preenrollment differences in this "Net" model are controlled for by including preenrollment observations in the equation and by estimating coefficients on preenrollment-calculated net-wage and nonwage-income variables. In the equation, coefficients a ₆ and a₇ measure preenrollment NIT differences (if any), and coefficients a₁ and a₂ measure the effect of the experiment. That is, the during-experiment dummy, D,is used to "net out" any preenrollment differences.
如上所述，在这个“净”模型中，通过在等式中包括入学前的观察值和估计入学前计算的净工资和非工资收入变量的系数来控制入学前差异。在等式中，系数 a ₆ 和 a₇ 衡量入组前的 NIT 差异（如果有），系数 a₁ 和 a₂ 衡量实验的效果。也就是说，实验期间的虚拟人 D 用于“净除”任何入组前差异。

Equation (4) is estimated with Tobit on a pooled sample of all seven during-experiment interviews and one preenrollment interview,including both attriters and those with changed marital status for the periods for which they have data available. Preenrollment wage rates are used (predicted for nonworkers). All these procedures we subjected to sensitivity tests, to be discussed in Section II.
方程式 （4） 是用 Tobit 对所有 7 次实验期间访谈和 1 次入组前访谈的汇总样本进行估计的，包括参与者和在他们有可用数据期间婚姻状况发生变化的人。使用预注册工资率（针对非工人的预测）。所有这些程序我们都经过了敏感性测试，将在第二节中讨论。

30

II. FINDINGS
二、研究结果

Estimates of α in equation (1) are shown in Table l. A range of estimates are provided by the use of two different statistical techniques to control for possible preenrollment experimental-control differences(see Appendix E). The table shows experimental effects on employment status and unconditional hours worked. The results show significant labor supply reductions for husbands and female heads, but not for wives. For husbands, employment-status reductions of 2.7 to 4.9 percent and uncon-ditional hours reductions of 2.9 to 6.5 percent occurred. For female heads, employment-status and unconditional-hours reductions of 26 to 30percent are found. The female-heads effects, it should be realized, are relative to a control group in which 80 percent of the female heads are on AFDC.
等式（1）中α的估计值见表l。通过使用两种不同的统计技术来控制可能的入组前实验对照差异，提供了一系列估计值（见附录 E）。该表显示了对就业状况和无条件工作时间的实验影响。结果显示，丈夫和女性头的劳动力供应显著减少，但妻子的劳动力供应却没有减少。对于丈夫来说，就业状况减少了2.7%至4.9%，非必要工作时间减少了2.9%至6.5%。对于女性负责人，就业状况和无条件工时减少了26%至30%。应该意识到，女性头部效应是相对于对照组而言的，其中80%的女性头部使用AFDC。

The results for husbands are very similar to those in other experiments, which ranged from l percent to 8 percent (Moffitt andKehrer, forthcoming). Consequently, this should increase our confidence in the general order of magnitude of the male NIT response. On the other hand, the results for wives are very different from those in the other experiments, which ranged from 15 percent to 55 percent (Moffitt andKehrer, forthcoming). One possible explanation is the very low employment rate of wives(15 percent): less than 90 of the 545 wives in the sample were working at preenrollment, experimental and control combined. This could have made differential experimental-control behavior difficult to detect statistically. This was probably a result of the Gary labor market,
丈夫的结果与其他实验的结果非常相似，范围从l%到8%不等（Moffitt和Kehrer，即将出版）。因此，这应该会增加我们对男性NIT反应的一般数量级的信心。另一方面，妻子的结果与其他实验的结果有很大不同，从15%到55%不等（Moffitt和Kehrer，即将出版）。一种可能的解释是妻子的就业率非常低（15%）：样本中的545名妻子中只有不到90名在入学前，实验和对照期间工作。这可能使差异实验控制行为难以在统计学上检测到。这可能是加里劳动力市场的结果，

31

which, as has been mentioned before, is dominated by the steel industry and provides very few part-time jobs.
如前所述，它以钢铁行业为主，提供的兼职工作很少。

The results for female heads are considerably larger than the12 percent response found in the Seattle-Denver experiment, the only other experiment designed to estimate the response of female heads. The absolute magnitudes of the hours reductions in the two experiments are close to one another but, given the much lower employment rate and hours worked in the Gary sample, a smaller response should be expected. The difference does not appear to be a result of different racial populations between the experiments because no statistically significant differences among the races were found in Seattle-Denver. Alternatively, the much lower AFDC standard in Indiana ($205 per month for a family of four in1972) compared to Seattle ($294) and Colorado ($242) could have made the net NIT stimulus (i. e., the NIT benefit minus the AFDC benefit) larger in Gary. However, the greater generosity of the Seattle-Denver NIT plans compensates for this difference. The net stimulus is about $100 per month for female heads in both experiments. Another possibility is simply that the tax and guarantee elasticities are stronger in Gary. To explore this hypothesis it is necessary to move to the estimates of equation (4).
女性头部的结果比西雅图-丹佛实验中发现的12%的反应要大得多，这是唯一一个旨在估计女性头部反应的其他实验。两个实验中工时减少的绝对幅度彼此接近，但考虑到加里样本的就业率和工作时间要低得多，预计响应应该会更小。这种差异似乎不是实验之间种族人口不同的结果，因为在西雅图-丹佛没有发现种族之间的统计学显着差异。或者，与西雅图（294美元）和科罗拉多州（242美元）相比，印第安纳州的AFDC标准（1972年四口之家每月205美元）要低得多，这可能使加里的净NIT刺激（即NIT福利减去AFDC福利）更大。然而，西雅图-丹佛NIT计划的慷慨弥补了这一差异。在这两个实验中，女性头部的净刺激约为每月100美元。另一种可能性是，加里的税收和担保弹性更强。为了探索这个假设，有必要转向方程（4）的估计。

A. BASIC MODEL
A. 基本型号

Table 2 shows the estimated budget-line coefficients of equation
表2显示了等式的估计预算项目系数

(4).¹/ F or husbands, the results show a weak income effect (significant
（4）.¹/F或丈夫，结果显示收入效应较弱（显著

l/Mean Tobit coefficients are presented because they are the relevant explanators of the total hours responses in Table 1. See McDonald and Moffitt (forthcoming). The original beta coefficient s can be obtained using the probabilities in the Table footnotes. Estimates of the other coefficients in the equation are available in Appendix E.
之所以列出 l/平均 Tobit 系数，是因为它们是表 1 中总小时响应的相关解释因素。见McDonald和Moffitt（即将出版）。原始 beta 系数 s 可以使用表脚注中的概率获得。等式中其他系数的估计值见附录 E。

32

at the 20-percent level) with a mean eelasticity of 5 percent, but an insignificant net-wage effect. These fairly inelastic responses are common for prime-age males and are not exceptional. They are also identical to the estimates ofBurtless and Hausman (1978), who only examined males and also found an insignificant net-wage elasticity and a significant income elasticity of 5 percent. Although there are other important model differences between this study and theirs, the most important is the different treatment cof the nonlinearity of the budget constraint. The conformity of result indicates that, at least on this sample, the average-tax-rate procedure is not a bad approximation. The conformity may again be a result of the highly structured Gary labor market. About 65 percent of the husbands worked in a steel mill and only 7 percent worked part-time (l to 34 hours per week)--the others worked around 38 hours per week or not at all. This lack of flexibility could make marginal hours adjustmentss difficult and therefore could make the averaging procedure-which throws away information on many intra-marginal tax rates--a satisfactory approximation. It should also be noted that this aspect of the Gary labor market may also explain why no net-wage (i. e., tax) effects were ffound here, as they have been in other experiments. Decreases in the ttax rate may not increase work effort if marginal hours increases cannot be made.
在20%的水平上），平均弹性为5%，但净工资效应微不足道。这些相当缺乏弹性的反应在壮年男性中很常见，并不例外。它们也与Burtless和Hausman（1978）的估计相同，他们只检查了男性，还发现净工资弹性微不足道，收入弹性为5%。尽管本研究与他们的研究之间存在其他重要的模型差异，但最重要的是对预算约束的非线性的不同处理。结果的一致性表明，至少在该样本上，平均税率程序不是一个糟糕的近似值。这种一致性可能再次是高度结构化的加里劳动力市场的结果。大约65%的丈夫在钢铁厂工作，只有7%的丈夫从事兼职工作（每周工作l至34小时），其他人每周工作约38小时或根本不工作。这种缺乏灵活性可能使边际工时调整变得困难，因此可能使平均程序（它丢弃了许多边际内税率的信息）成为一个令人满意的近似值。还应该指出的是，加里劳动力市场的这一方面也可以解释为什么这里没有发现净工资（即税收）效应，就像在其他实验中一样。如果不能增加边际工时，税率的降低可能不会增加工作量。

The results for wives show both insignificant net-wage and insignificant income effects. This confirms the total lack of response discussed above. Again, the low employment rate (15 percent) of the sample may be part of the explanation, together with the occupational structure of the local labor market.Another indication of this constraint
对妻子的调查结果显示，净工资效应微不足道，收入效应微不足道。这证实了上文讨论的完全缺乏答复。同样，样本的低就业率（15%）可能是解释的一部分，以及当地劳动力市场的职业结构。这种限制的另一个迹象

33

is that part-time work was very infrequent--9 percent of the sample-compared to the amount of part-time work among wives in other samples.
与其他样本中妻子的兼职工作数量相比，兼职工作非常少 - 占样本的9%。

For female heads, the response shows a significantly negative income effect with an elasticity of .23, but an insignificant net-wage effect. The lack of a response to the tax rate may again be a result of inflexibility in the Gary labor market. But, like husbands, female heads did respond to the guarantee. Indeed, the magnitudes of the husbands'and female heads' coefficients are rather close to one another. This may be because female heads also must often provide earnings for an entire family, similar to many husbands.
对于女性头，响应显示显着的负收入效应，弹性为0.23，但净工资效应微不足道。对税率缺乏回应可能再次是加里劳动力市场缺乏灵活性的结果。但是，像丈夫一样，女性首领确实对这一保证做出了回应。事实上，丈夫和女性头部的系数大小相当接近。这可能是因为女性户主也必须经常为整个家庭提供收入，类似于许多丈夫。

The size of the female-head guarantee effect goes a long way twoard explaining why the response here is larger than that in Seattle-Denver. The comparable income effect in Seattle-Denver is less than one-third of the Gary effect (-2.02 vs. -6.35). The effect of this difference on the response can be illustrated by applying the Seattle-Denver coefficients to the Gary guarantee-and-tax-rate stimuli. When this is done, the predicted Gary response of 26-30 percent falls by one-half(i. e., to 13-15 percent). This explains most of the difference with theSeattle-Denver 12-percent response.
女性头部保证效应的大小在很大程度上解释了为什么这里的反应比西雅图-丹佛的反应更大。西雅图-丹佛的可比收入效应不到加里效应的三分之一（-2.02 vs. -6.35）。这种差异对响应的影响可以通过将西雅图-丹佛系数应用于加里保证和税率刺激来说明。当这样做时，预测的26-30%的加里反应下降了一半（即，下降到13-15%）。这解释了与西雅图-丹佛 12% 的回应的大部分差异。

Why the Gary female heads have a stronger income effect is unclear. The most likely explanation lies in the fact that the Seattle-Denver sample of female heads is fairly high-income, at least compared to that in Gary. For this reason Seattle-Denver female heads may simply have had a greater committed level of work effort. For example, Seattle-Denver female heads worked an average of 80 hours per month whereas
为什么加里女性头具有更强的收入效应尚不清楚。最可能的解释在于，西雅图-丹佛的女性负责人样本收入相当高，至少与加里相比是这样。出于这个原因，西雅图-丹佛的女性领导可能只是付出了更大的工作努力。例如，西雅图-丹佛的女性负责人平均每月工作 80 小时，而

34

Gary female heads worked only 31 hours. If there is a nonlinearity in the response to an NIT, with greater responses at lower income and hours levels, this could explain the difference.
加里女头只工作了 31 个小时。如果对NIT的响应存在非线性，那么在较低的收入和工时水平上，响应更大，这可以解释这种差异。

Table 2 also shows the coefficients on the non-NIT income variables variables. The most striking difference is in the magnitude and/or significance of the NIT and non-NIT coefficients. Relative to the NITcoefficients, the non-NIT coefficient s show more significant and larger income effects for husbands, significant net-wage effects for wives, and more significant income effect for female heads. (However, wives have a significant but positive non-NIT income effect.) Moreover, the coefficient differences are significant at the 5 percent level. These results show the importance of estimating NIT tax and guarantee coefficients separately from those of non-NIT income. This could also further explain some of the above-mentioned differences between the NIT estimates in this study and those in other experimental studies where the coefficients were constrained to be identical.
表2还显示了非NIT收入变量的系数。最显着的区别在于 NIT 系数和非 NIT 系数的大小和/或显着性。与NITcoefficients相比，非NIT系数对丈夫的收入效应更显著且更大，对妻子的净工资效应更显著，对女性户主的收入效应更显著。（然而，妻子对非NIT收入有显著但积极的影响。此外，在5%的水平上，系数差异是显著的。这些结果表明，将NIT税收和担保系数与非NIT收入分开估算具有重要意义。这也可以进一步解释本研究中的NIT估计值与其他实验研究中的NIT估计值之间的一些上述差异，在这些实验研究中，系数被限制为相同。

Numerous sensitivity tests were performed on this equation. The equation was estimated with OLS and generalized least squares, instead of Tobit; preenrollment differences were controlled for by lagging the dependent variable instead of "netting out" preenrollment treatment differences; the data were averaged instead of pooled; only the middle year of the experiment was examined; a variety of wage instruments were used; and sc on. We discuss these in the next section. The results show that the estimates of the basic equation are quite robust,especially given the insignificance of many of the coefficients. Perhaps most notable is that non e of these tests revealed any hidden response of wives.
对该方程式进行了许多灵敏度测试。该方程是使用 OLS 和广义最小二乘法估计的，而不是使用 Tobit;通过滞后因变量而不是“剔除”入组前治疗差异来控制入组前差异;数据是平均的，而不是合并的;只检查了实验的年中;使用了各种工资工具;和 SC 上。我们将在下一节中讨论这些内容。结果表明，考虑到许多系数的不显著性，基本方程的估计具有较强的鲁棒性。也许最值得注意的是，这些测试中没有一项揭示妻子的任何隐藏反应。

35

TABLE 1
表1

ESTIMATED MEAN EXPERIMENTAL-CONTROL LABOR SUPPLY DIFFERENCES
估计平均实验对照劳动力供给差异

a/See Appendix E for full regression results.
a/有关完整的回归结果，请参阅附录 E。

b/No significance tests. See Appendix E.
b/无显著性检验。见附录E。

**Coefficient significant at 95 percent level.
**系数在 95% 水平下显著。

***Coefficient significant at 99 percent level.
系数在99%水平上显著。

TABLE 2
表2

MEAN COEFFICTENTS IN TOBIT HOURS/MONTH REGRESSIONR/
以 TOBIT 小时/月回归为单位的平均系数

Notes: Unsigned t-statistics are in parentheses.
注意：无符号的 t 统计量在括号中。

*Significant at the 10-percent level.
*在10%的水平上很重要。

**Significant at the 5-percent level.
**在 5% 的水平上显着。

***Signifiaant at the l-percent level.
在 l% 水平上具有显著性。

a/All coefficients evaluated at the mean value of the Tobit index. That is, if β is the vector of coefficients, X is the vector of independent variables, and F is the cumulative normal distribution function, the coefficients in the table are β F (Xβ/σ). F(Xβ/σ) equals .99, .15, and.2l for husbands, wives, and female heads, respectively.
a/所有系数均以 Tobit 指数的平均值计算。也就是说，如果 β 是系数的向量，X 是自变量的向量，F 是累积正态分布函数，则表中的系数β F （Xβ/σ）。F（Xβ/σ） 分别等于 .99、.15 和 .2l 的丈夫、妻子和女性头。

b/Divided by 100.
b/除以100。

B. SENSITIVITY TESTS
B. 敏感性测试

1. Dependent Variable
1. 因变量

Table 3 shows the results of the various sensitivity tests, the first of which involves the use of a different dependent variable. The other dependent variable examined is employment status, defined as equal to one if the person was employed 50% of the month or more, and zero otherwise. As the probit-estimated coefficients in the table show, the pattern of the signs and significance level is the same for employment status as for hours. Furthermore, when the probit coefficients are multiplied by mean hours, the resulting implied hours effects are quite close to the Tobit hours effects (at least in those cases where the underlying coefficients. . are significant). This provides direct evidence that most of the response occurred in employment status and indirect evidence that hours of work are inflexible in the Gary labor market.
表 3 显示了各种敏感性测试的结果，其中第一个涉及使用不同的因变量。检查的另一个因变量是就业状况，如果该人每月受雇 50% 或更多，则定义为等于 1，否则为 0。如表中的概率估计系数所示，就业状况的标志和显著性水平的模式与小时数相同。此外，当概率系数乘以平均小时数时，得到的隐含小时数效应非常接近托比特小时数效应（至少在基础系数显著的情况下）。这提供了直接证据，表明大多数反应发生在就业状况上，间接证据表明加里劳动力市场的工作时间不灵活。

2. Estimating Technique
2. 估算技术

The basic model for hours worked is also estimated with ordinary least squares (OLS); the guarantee and net-wage coefficients are shown in Table3. These estimates are very close to the (mean) Tobit estimates. Also,the employment status equations have been estimated both by OLS and by a generalized least squares (GLS) method designed to control for the cross-section time-series nature of the data set (it is an error-component model with a constant individual effect and a non-auto-correlated random effect). The table shows. that the OLS estimates are very similar to the probit estimates. The GLS estimates are also close to the probit estimates
工作时间的基本模型也是用普通最小二乘法 （OLS） 估计的;保障和净工资系数见表3。这些估计值非常接近（平均值）Tobit 估计值。此外，就业状态方程已通过OLS和广义最小二乘法（GLS）方法进行估计，该方法旨在控制数据集的横截面时间序列性质（它是一种具有恒定个体效应和非自相关随机效应的误差分量模型）。该表显示。OLS估计值与probit估计值非常相似。GLS的估计值也接近概率估计值

38

TAELE 3
泰乐 3

SUMMARY OF SENSITIVITY TESTS ON AVERAGE-TAK-RATE MODEL
AVERAGE-TAK-RATE模型的敏感性检验总结

Husbands WivesFemale Heads
丈夫 妻子女性 头

Guarantee Net-Wage Guarantee Net-Wage Guarancee Net-wage
保证净工资 保证净工资 保证净工资

Ccefficient Coefficient Coefficient Coefficient :Coefficient Coefficient
Ccefficient Coefficient Coefficient ：系数系数

1. Dependent variablea/
1. 因变量a/

. H (Tobit) -5.85 -3.60 4.07 12.67 -6.35* -4.61
.H （托比特） -5.85 -3.60 4.07 12.67 -6.35* -4.61

(1.42) (.36) (.58) (.63) (1.91) (.49)

ES (probit) -.049* -.057 .38 .066 -.040* -.02
ES（概率） -.049* -.057 .38 .066 -.040* -.02

(1.31) (.89) (1.47) (.86) (1.a3) (.35)
（1.31） （.89） （1.47） （.86） （1.a3） （.35）

Hours effects derived from ES equation
从ES方程推导出的小时效应

2. Estimating techniqueb/ -5.46
2. 估算技术b/ -5.46

H (OLS) -5.46 -3.68 4.15 5.15 -6.45* -4.03
H （OLS） -5.46 -3.68 4.15 5.15 -6.45* -4.03

(1.47) (.41) (.67) (.27) (1.83) (.39)

ES (OLS) -.034 -.032 .037 .109 -.032 .002
ES （OLS） -.034 -.032 .037 .109 -.032 .002

(1.56) {.62} (.40) (.86) (1.45) (.03)

ES (GLS) -.031** -.045 .036 .031 -.051** -.072*
ES （GLS） -.031** -.045 .036 .031 -.051** -.072*

{2.09} (1.28) (1.13) (.33) (3.61) (1.93)

3 with period dumnies(Tobit)
3 带周期假人（Tobit）

Income underreporting(OLS)ミ'
收入少报（OLS）ミ'

"validated" ES -.032 -.013 -.007 .027 -.041* -.076
“已验证” ES -.032 -.013 -.007 .027 -.041* -.076

(1.54) (.31) (.14) (.19) (1.79) (1.13)

Sample Selection
样品选择

?.そ3ddition to sample of screened-out groups(0Ls)ミﺑ
?.そ3ddition 对筛选出的组别样品（0Ls）ミب

H -4.63 -5.54 -7.58 -19.57 -6.24 -3.20

(1.20) (.16) (.89) (1.01) (1.58) (.30)

ES -.033 -.048 .034 .113 ~.017 .023
欧洲杯 -.033 -.048 .034 .113 ~.017 .023

(1.40) (.84) (.85) (.95) (.85) (.36)

Delet ion from sample of attriters and discon-tinus families?
从Attriters和Discon-tinus家族的样品中去除离子？

H (Tobit) -5.67 -2.47 -4.03 -14.53 -8.40** -10.12
H（托比特） -5.67 -2.47 -4.03 -14.53 -8.40** -10.12

(1.25) (.22) (.50) (.60) (2.54) (1.10)

Preenrollment differencesess/
入学前差异/

E (Tobit) -3.81** -5.47 -6.67** -14.72* -4.50** -4.39
E（托比特） -3.81** -5.47 -6.67** -14.72* -4.50** -4.39

(1.98) (1.21) (2.21) (1.78) (3.8) (1.59)

ES (OLS) -.016 -.012 .003 .012 -.023** -.033
ES （OLS） -.016 -.012 .003 .012 -.023** -.033

(1.58) (.49) (.12) (.19) (2.92) (1.35)

う. Time period
う. 时间段

Middle Year OnlyE (Tobit)
中学 OnlyE （Tobit）

Middle Year AveragedH (Tobit)
初中平均dH （Tobit）

39

TES: All guarantee coefficients multiplied by 100. Unsigned t-statistics in parentheses.
TES：所有保证系数乘以 100。括号中的无符号 t 统计量。

*Significant at the 90-percent level.
*在 90% 的水平上很重要。

**Significant at the 95-percent level.
**在 95% 的水平上很重要。

H = hours per month
H = 每月小时数

ES = employment status (1 if employed 50 percent of month, 0 otherwise)
ES = 就业状况（如果每月就业率为 50%，则为 1，否则为 0）

OLS = ordinary least squares
OLS = 普通最小二乘法

a Tobit and probit coefficients evaluated at mean of index, ES hours effects obtained by multiplying ES
a Tobit系数和Probit系数以指数的平均值评估，通过乘以ES得到的ES小时效应

coefficients by the difference in mean hours of those with ES = 1 and ES = 0.
ES = 1 和 ES = 0 时的平均小时差的系数。

S. GLS (generalized least squares) is an error-component tech nique with a constant individual effect and
S. GLS（广义最小二乘法）是一种具有恒定个体效应的误差分量技术，并且

a non-auto-correlated random effect.
非自相关随机效应。

These equations were estimated by Caroline Roth.
这些方程是由卡罗琳·罗斯（Caroline Roth）估计的。

ed/Readdition of self-employed, aged , young, and disabled. These equations were estimated by David Homer.
ed/重新增加个体经营者、老年人、年轻人和残疾人。这些方程是由大卫·荷马（David Homer）估计的。

E/The "Lagged" model is used: only during-experiment observations are included, and the preenrollment
E/使用“滞后”模型：仅包括实验期间的观察结果，并且预注册

value of hours worked is included on the right-hand-side.
工作时间值包含在右侧。

1. The "mixed preenrollment wage" is constructed by using actual preenrollment wages for workers and a eenrollment wage instrument for nonworkers. The "pure preenrollment instrument" is constructed by using the preenrollment w age instrument for all observations. The "mixed contemporaneous wage" is constructed by using the actual during-experiment wage for workers and a during-experiment wage instrument for nonworkers. The "pure"ntemporaneous instrument" is constructed by using the during-experiment wage instrument for all observations.
1. “混合预登记工资”是使用工人的实际预登记工资和非工人的电子登记工资工具构建的。“纯入组前工具”是通过使用所有观察的入组前 w 年龄工具构建的。“混合同期工资”是利用工人的实际实验期工资和非工人的实验期工资工具构建的。“纯”时间仪器“是通过使用实验期间工资仪器进行所有观察而构建的。

40

except for female heads, where both coefficients are more negative; in fact, the compensated substitution effect is negative, an implausible result. Finally, the addition of period dummies to remove period effects from the pooled data also makes very little impact on the estimates.
除了女性头部，两个系数都更为负;事实上，补偿替代效应是负面的，这是一个难以置信的结果。最后，添加周期假人以从合并数据中去除周期效应对估计值的影响也很小。

3. Income Underreporting
3. 收入少报

The test using a "validated" employment-status dependent variable(i. e., one collected from government records) shows little effect of underreporting, for the coefficients are quite similar to those in theES regressions using only interview data. However, it should be noted that Greenberg et al. (1979) have used this same validated data to investigate the question in more de tail and have come to a rather different conclusion, at least for female heads. Therefore the effect of underreporting on the response is unclear. See the Greenberg et al.paper for details.
使用“经过验证的”就业状况因变量（即从政府记录中收集的变量）的检验显示漏报的影响很小，因为系数与仅使用访谈数据的ES回归中的系数非常相似。然而，应该指出的是，Greenberg等人（1979）使用相同的验证数据来研究这个问题，并得出了相当不同的结论，至少对于女性头部而言。因此，报告不足对应对措施的影响尚不清楚。有关详细信息，请参阅 Greenberg 等人的论文。

4. Sample Selection
4. 样品选择

The table also shows that the readdition of the screened-out groups (aged, young, self-employed, and disabled) has little effect on the estimated coefficients for husbands and female heads. The coefficients in the wives' hours equation are, however, considerably different, giving negative income effects and negative compensated substitution effects.Given the perversity of the latter and the low significance levels, little weight should be given to this result.
该表还显示，重新增加被筛选出的群体（老年人、年轻人、自营职业者和残疾人）对丈夫和女性户主的估计系数影响不大。然而，妻子工时方程式中的系数有很大不同，产生了负的收入效应和负的补偿替代效应。鉴于后者的反常性和低显著性水平，不应对这一结果给予很少的重视。

The Tobit hours equation was also reestimated with the attriters and discontinuous families deleted. The two groups are simultaneously
还重新估计了Tobit小时方程，删除了属性体和不连续族。这两个组同时是

41

deleted because of the difficulty in distinguishing them (e. g., many husband-wife families may leave the experiment as they split up). The estimates in the table show that the treatment coefficients for husbands and female heads are rather stable (although female heads have more negative income and substitution effects) and that the coefficients for wives are again unstable but highly insignificant statistically.
由于难以区分它们而被删除（例如，许多夫妻家庭可能会在分手时退出实验）。表中的估计数表明，丈夫和女性户主的待遇系数相当稳定（尽管女性户主的负收入和替代效应更多），妻子的系数同样不稳定，但在统计上非常微不足道。

5. Preenrollment Differences
5. 入学前差异

As indicated in section II-C above, the alternate method used to control for preenrollment differences is the estimation of the "Lagged"model. In this model only during-experiment observations are used and the preenrollment value of the dependent variable is included on the right-hand-side of the labor-supply equation. Both a Tobit hours model and an OLS employment-status model have been estimated, and the results are shown in the table. Once again, the treatment coefficients for husbands and female heads are relatively stable, although the effects on the employment status of husbands are smaller in the Lagged model.However, the results for wives are again very unstable. In the hours equation the wives' coefficients show a significant and negative guarantee effect but also a signfican t and negative net-wage effect (the compensated substitution effect is also negative). This anomalous result may be caused by the lagged dependent variable, which may bias the other coefficients in the equation because the lagged value contains measurement error and because it is itself a function of the wage rate, nonwage income, and so on. On the other hand, the employment-status results for wives show even
如上文第II-C节所述，用于控制入学前差异的替代方法是“滞后”模型的估计。在该模型中，仅使用实验期间的观察值，并将因变量的预注册值包含在劳动力供给方程的右侧。对Tobit小时数模型和OLS就业状况模型都进行了估计，结果如表所示。同样，丈夫和女性户主的待遇系数相对稳定，尽管在滞后模型中对丈夫就业状况的影响较小。然而，妻子的结果又非常不稳定。在小时方程中，妻子系数表现出显著的负保证效应，但也表现出显著的负的净工资效应（补偿替代效应也是负的）。这种异常结果可能是由滞后因变量引起的，这可能会使方程中的其他系数产生偏差，因为滞后值包含测量误差，并且因为它本身就是工资率、非工资收入等的函数。另一方面，妻子的就业状况结果显示均匀

42

smaller and less significant coefficients in the Lagged model compared to the Net model. Therefore once again we find no evidence of a sensible wives'response.
与 Net 模型相比，Lagged 模型中的系数较小且不显著。因此，我们再一次没有发现任何证据表明妻子会做出明智的反应。

6. Time Period
6. 时间段

Estimation of the model on the middle year's data, whether by pooling the monthly observations or by averaging them, has very little impact. For husbands and female heads, income effects are somewhat stronger but substitution effects are correspondingly weaker. Changing the time period of the analysis by using quarterly data instead of monthly data and by averaging our observations rather than pooling also has little effect on the husbands' income coefficient, and little effect on the insignificance of the husbands' net-wage coefficient or either of the wives' coefficients. For female heads, quarterly pooled estimation makes little difference but averaging our periods forces both more negative income effects and more negative substitution effects. No ready explanation is apparent for this shift.
对中年数据的模型进行估计，无论是通过汇总月度观测值还是平均值，影响都很小。对于丈夫和女性首领来说，收入效应稍强，但替代效应相应较弱。通过使用季度数据而不是月度数据来改变分析的时间段，并通过平均我们的观察结果而不是合并来改变分析的时间段，对丈夫的收入系数也没有什么影响，对丈夫的净工资系数或妻子的任何一个系数的微不足道也没有什么影响。对于女性负责人来说，季度汇总估计几乎没有区别，但平均我们的经期会带来更多的负面收入效应和更多的负面替代效应。对于这种转变，没有现成的解释。

7. Wage Rate
7. 工资率

The use of different wage rates has remarkably little effect on the coefficient estimates (the female heads' net-wage coefficient being the one exception). The lack of an important difference between the estimates using preenrollment and contemporaneous wage rates is no doubt a result of the lack of significant effects of the experiment on wage rates. But also, our separation of the net wage into NIT and non-NITcomponents eliminates to some degree the importance of the wage rate in affecting the NIT coefficients--that is, (-Wt) varies independently of Wonly through variations in t.
使用不同的工资率对系数估计的影响非常小（女校长的净工资系数是一个例外）。使用入学前工资率和同时期工资率的估计值之间缺乏重要差异，这无疑是实验对工资率缺乏重大影响的结果。而且，我们将净工资分为NIT和非NIT成分，在一定程度上消除了工资率在影响NIT系数方面的重要性，也就是说，（-Wt）通过t的变化独立于Wonly而变化。

43

8. Household Interdependence
8. 家庭相互依存

The use of actual, contemporaneous husbands' earnings in the nonwage-income variable for wives (rather than an instrument) does increase the net-wage effect and is significant. However, the income effect is still positive and a bit larger and more significant. Therefore this very simple test of the importance of household interdependence provides mixed results.
在妻子的非工资收入变量中使用实际的、同期的丈夫收入（而不是一种工具）确实增加了净工资效应，而且是显著的。然而，收入效应仍然是积极的，而且规模更大、更显著。因此，这种对家庭相互依存重要性的非常简单的检验提供了不同的结果。

9. Other Tests
9. 其他测试

Other sensitivity tests were also performed which are not shown in the table. Addition of the guarantee and tax rate (actually, the net wage difference) for food stamps had no effect on the estimated coefficients.Also, estimation of a "local tax rate" model similar to that of Keeley et al. (1978) provided mixed results, showing some stronger effects for some groups but more perverse effects for others. Details of this test are discussed in Appendix G.
还进行了表中未显示的其他敏感性测试。加上食品券的担保和税率（实际上是净工资差额）对估计系数没有影响。此外，对类似于Keeley等人（1978年）的“地方税率”模型的估计结果喜忧参半，对某些群体显示出更强的影响，但对其他群体的影响更差。该测试的详细信息在附录 G 中讨论。

44

エエエ. SUMMARY
赞成。 总结

The results we have found for the Gary NIT experiment show work disincentives for some groups but not for others. In particular,significant responses by prime-age married males ranged from 2.9 percent to 6.5 percent of hours worked, and those for female heads of families ranged from 25.9 percent to 30 percent. In addition, these responses seem to be primarily responses to the guarantee level rather than to the tax rate. On the other hand, no response at all was found for wives, either in the aggregate or by tax rate and guarantee level.
我们在Gary NIT实验中发现的结果显示，某些群体的工作抑制作用，但对其他群体则没有。特别是，壮年已婚男性的显著反应范围为工作时间的2.9%至6.5%，女性户主的反应范围为25.9%至30%。此外，这些反应似乎主要是对担保水平的反应，而不是对税率的反应。另一方面，无论是在总体上还是在税率和担保水平上，都没有发现妻子的回应。

These findings require some interpretation. For husbands, at least, the findings are quite similar to those in other experiments,although a bit smaller. The finding of a guarantee response but not a tax-rate response may be a result of the inflexibility of the Gary labor market, as discussed in the paper. This hypothesis is reinforced by the seemingly larger response in employment status than in hours of work.For female heads, the response is a good bit larger than that in theSeattle-Denver experiment, where a negative response was also found but was considerably smaller in percentage terms. O ur feeling is that this is largely a result of differences in the characteristics of the samples,for the female heads in the Gary experiment were lower in income and more dependent on AFDC than those in the Seattle-Denver experiment.
这些发现需要一些解释。至少对于丈夫来说，这些发现与其他实验中的发现非常相似，尽管要小一些。正如本文所讨论的，发现担保响应而不是税率响应可能是加里劳动力市场缺乏灵活性的结果。就业状况的反应似乎大于工作时间，这强化了这一假设。对于女性头部，反应比西雅图-丹佛实验中的反应要大得多，在西雅图-丹佛实验中也发现了负面反应，但百分比要小得多。哦，你的感觉是，这主要是样本特征差异的结果，因为加里实验中的女性头部收入较低，并且比西雅图-丹佛实验中的女性头部更依赖AFDC。

The findings for wives are the most dissimilar to those in other experiments, where it was often found that wives experienced the greatest work disincentives of any group. We believe that, in part, this is a result
妻子的发现与其他实验的结果最不同，在其他实验中，人们经常发现妻子经历了任何群体中最大的工作抑制。我们认为，这在一定程度上是一个结果

45

of the nature of the Gary labor market, which offered few work opportunities for females. In fact, the employment rate for wives in our sample was very low (15 percent). In part, it also is a result of the types of working women in the sample, many of whom were full-time,career-oriented workers rather than part-time, casual workers. Although in the final analysis these factors are not enough to expect a zero response, they do provide reason to expect it to be lower than that in other experiments.
加里劳动力市场的性质，为女性提供的工作机会很少。事实上，在我们的样本中，妻子的就业率非常低（15%）。在某种程度上，这也是样本中职业女性类型的结果，其中许多人是全职的、以职业为导向的工人，而不是兼职的临时工。尽管归根结底，这些因素不足以预期零响应，但它们确实提供了预期它低于其他实验中的理由。

The results also show that the measured effects of NIT and non-NIT income are very different. The differences may be explained by limited-duration biases; some evidence (Moffitt, 1979) suggests that it might be. Alternatively, the differences may be true differences--i. e.,people may not actually respond identically to different income sources and tax-and-transfer programs--or they may be spurious, simply resulting from inherent biases in nonexperimental data. The nature of the difference suggests an important area for future research.
结果还表明，NIT和非NIT收入的测量效果非常不同。这些差异可以用有限持续时间的偏倚来解释;一些证据（Moffitt，1979）表明它可能是。或者，差异可能是真正的差异--i。例如，人们实际上可能不会对不同的收入来源和税收和转移支付计划做出相同的反应 - 或者它们可能是虚假的，仅仅是由于非实验数据中的固有偏见造成的。这种差异的性质表明了未来研究的一个重要领域。

46

APPENDIX A
附录A

SAMPLE CHARACTERISTICS
样品特性

47

TABLE A-1
表 A-1

MEAN CHARACTERISTICS OF HUSBANDS
丈夫的平均特征

Preenrollment During-Experiment
预注册期间-实验

Industry of work (percentage distribution);
工作行业（百分比分布）;

Construction
建设

Non-steel manufacturing
非钢制造业

Steel manufacturing
钢铁制造

Transportation, Communi--
交通、通讯--

cation, and Utilities
阳离子和公用事业

Wholesaling and retailing
批发和零售

Finance, insurance, and
金融、保险和

real estate
房地产

Services
服务业

Government
政府

Otherd/
其他/

Occupation of work (per-ce ntage distribution):
工作职业（per-ce ntage distribution）：

Professional and
专业和

technical
专门的

Managers
经理

Sales
销售

Clerical
文书

Craftsmen
工匠

Non-transport services
非运输服务

Transport services
运输服务

Laborers
劳动者

Service workers, exp.household
服务人员，exp.household

Private household
私人家庭

Other
其他

Age
年龄

Years of Education
教育年限

T/DI-E A--Lcontinued)
T/DI-E A--L续）

Preenrollment During-Experiment
预注册期间-实验

Experimental Control Experimental Control
实验对照 实验对照

Group Group Total Group Group Total
集团合计 集团合计

Regression variables:
回归变量：

No. adultse/ 2.5 2.5 2.5 2.6 2.5 2.5
不。成人/ 2.5 2.5 2.5 2.6 2.5 2.5

No. children 3.2 3.1 3.2 3.4 3.2 3.4
不。儿童 3.2 3.1 3.2 3.4 3.2 3.4

Presence of children:
有儿童在场：

0-2 .34 .26 .31 .35 .26 .31

3-5 .41 .43 .42 .39 .44 .41

6-12 .69 .69 .69 .73 .74 .73

13-15 .47 .45 .46 .52 .46 .49

Multiple family .05 .04 .05 .02 .03 .03
多族 .05 .04 .05 .02 .03 .03

Season .75 .83 .79 .31 .31 .31
季节 .75 .83 .79 .31 .31 .31

Unemployment rate 3.56 3.48 3.52 4.80 4.82 4.80
失业率 3.56 3.48 3.52 4.80 4.82 4.80

E /l00o

-1.25 0 -.70 -1.16 0 -.70(-Wt)
-1.25 0 -.70 -1.16 0 -.70（-重量）

B。/100-actual 0 0 0 2.76 0 1.60
B。/100-实际值 0 0 0 2.76 0 1.60

(-:Wt)-actual. 0 0 0 -1.16 0 -.70
（-：Wt）-实际。0 0 0 -1.16 0 -.70

N/l00 .72 .75 .73 .72 .79 .75

W(l-r) 3.12 3.12 3.12 3.14 3.16 3.15
宽（从左到右） 3.12 3.12 3.12 3.14 3.16 3.15

Number of individuals in sample
样本中的个体数

NOTES:
笔记：

a/₁Unemployment = l if unemployed, 0 if employed or not in labor force.
a/₁失业 = l（如果失业），0（如果就业或未从事劳动力）。

b/Less than 0.005 in absolute value.
b/绝对值小于0.005。

c/Zeros included
c/零

d/Includes non-workers
d/包括非工人

e/Includes principal person.
e/包括主要人员。

TABLE A-2
表 A-2

MEAN CHARACTERISTICS OF WIVES
妻子的平均特征

Preenrollment
预注册

-MABLE-n-2 (pnm*inuéd...
-MABLE-n-2 （pnm*inuéd...

Preenrollment During-Experiment
预注册期间-实验

Experimental Control Experimental Control
实验对照 实验对照

Group Group Total Group Group Total
集团合计 集团合计

Regcession variables:
Regcession 变量：

No. adultse/ 2.6 2.5 2.5 2.6 2.5 2.6
不。成人/ 2.6 2.5 2.5 2.6 2.5 2.6

No. children 3.1 3.1 3.1 3.4 3.2 3.3
不。儿童 3.1 3.1 3.1 3.4 3.2 3.3

Presence of children:
有儿童在场：

0-2 .33 .26 .30 .34 .25 .31

3-5 .40 .42 .40 .38 .43 .40

6-12 .68 .69 .69 .72 .73 .72

13-15 .47 .45 .46 .52 .46 .49

Multiple family .05 .05 .05 .02 .03 .03
多族 .05 .05 .05 .02 .03 .03

Season .76 .83 .79 .32 .31 .32
季节 .76 .83 .79 .32 .31 .32

Unemployment rate 3.56 3.47 3.52 4.80 4.82 4.81
失业率 3.56 3.47 3.52 4.80 4.82 4.81

E./100 .99 .0 .55 .69 .0 .41

(-Wt) -.33 0 -.18 -.26 0 -.15
（-重量） -.33 0 -.18 -.26 0 -.15

B。/100-actual 0 0 0 .69 0 .41
B。/100-实际 0 0 0 .69 0 .41

(-Wt)-actual 0 0 0 -.26 0 -.15
（-Wt）-实际 0 0 0 -.26 0 -.15

N/l00 5.66 5.78 5.71 5.82 5.96 5.88

W(l-r) 1.79 1.72 1.76 1.76 1.71 1.74
宽（左-右） 1.79 1.72 1.76 1.76 1.71 1.74

Number of individuals in sample
样本中的个体数

NOTES:.
笔记：。

a/Unemployment = l if unemployed, 0 if employed or not in labor force
a/失业 = l（如果失业），0（如果就业或未就业）

b/Less than 0.005 in absolute value
b/绝对值小于0.005

C/zeros included
包括 C/零

d/Includes nonworkers
d/包括非工人

e/Includes principal person.
e/包括主要人员。

TABLE A-3
表 A-3

MEAN CHARACTERISTICS OF FEMALE HEADS
女性头部的平均特征

Preenrollment During-Experiment
预注册期间-实验

Experimental Control Experimental Control
实验对照 实验对照

Group Group Total Group Group Total
集团合计 集团合计

Industry of work (percentage distribution);
工作行业（百分比分布）;

Construction 一 一 一
建筑 一 一

Non-steal manufacturing .03 .04 .03 .02 .02 .02
非偷窃制造 .03 .04 .03 .02 .02 .02

Steel manufacturing .02 .03 .02 .02 .03 .02
钢铁制造 .02 .03 .02 .02 .03 .02

Transportation, Communi-cation, and Utilities
交通、通信和公用事业

Wholesaling and retailing .10 .05 .08 .04 .04 .04
批发和零售 .10 .05 .08 .04 .04 .04

Finance, insurance, and real estate
金融、保险和房地产

Services .15 .14 .15 .08 .10 .09
服务 .15 .14 .15 .08 .10 .09

Government .03 .04 .03 .02 .03 .02
政府 .03 .04 .03 .02 .03 .02

Other r:60 1:710 т:50 1:82 1:79T:810
其他 r：60 1：710 т：50 1：82 1：79T：810

Occupation of work (per-ce ntage distribution);
工作职业（per-ce ntage distribution）;

Professional and technical
专业技术

Managers 一 一 - … - -
经理 一 一 - ... - -

Sales 一 ~ 一 - - -

Clerical .09 .06 .08 .06 .04 .05
文员 .09 .06 .08 .06 .04 .05

- .01 一 - 一 →Craftsmen
- .01 一 - 一 →工匠

Non~transport services .05 .05 .05 .02 .03 .03
非~运输服务 .05 .05 .05 .02 .03 .03

Transport operatives
运输人员

Laborers .02 .01 .02 .01 .02 .01
劳动者 .02 .01 .02 .01 .02 .01

Service workers, exp.household
服务人员，exp.household

Private household .02 .02 .01 .01
私人家庭 .02 .02 .01 .01

Other 1:00.65 1:68 1:66 1:79 .95 1:79
其他 1：00.65 1：68 1：66 1：79 .95 1：79

Age 36.0 34.9 35.6 36.7 36.2 36.5
年龄 36.0 34.9 35.6 36.7 36.2 36.5

Years of Education 10.2 10.3 10.2 10.2 10.4 10.3
教育年限 10.2 10.3 10.2 10.2 10.4 10.3

Tanx- E A-3--conduced)
Tanx- E A-3--导致）

Preenrollment During-Experiment
预注册期间-实验

Experimental Control Experimental Control
实验对照 实验对照

Group Group Total Group Group Total
集团合计 集团合计

Regression variables:
回归变量：

No. adultse/ 1.5 1.4 1.4 1.6 1.4 1.5
不。成人/ 1.5 1.4 1.4 1.6 1.4 1.5

No. children 2.8 3.0 2.9 2.7 3.0 2.8
不。儿童 2.8 3.0 2.9 2.7 3.0 2.8

Presence of children:
有儿童在场：

0-2 .21 .25 .23 .27 .31 .29

3~5 .32 .36 .34 .31 .36 .33

6-12 .62 .69 .65 .59 .66 .62

13-15 .40 .46 .43 .37 .43 .40

Multiple family .07 .05 .06 .09 .06 .08
多家庭 .07 .05 .06 .09 .06 .08

Season .56 .62 .59 .29 .30 .29
赛季 .56 .62 .59 .29 .30 .29

Unemployment rate 3.88 3.78 3.84 4.82 4.84 4.83
失业率 3.88 3.78 3.84 4.82 4.84 4.83

Eo/100 2.50 0 1.47 2.18 0 1.31
EO/100号 2.50 0 1.47 2.18 0 1.31

(-Wt) -.84 0 -.49 -.80 0 -.48
（-重量） -.84 0 -.49 -.80 0 -.48

Bo/l00-actual 0 0 0 2.18 0 1.31
Bo/l00-实际值 0 0 0 2.18 0 1.31

(--Wt)-actual 0 0 0 -.80 0 -.48
（--Wt）-实际值 0 0 0 -.80 0 -.48

N/100 1.93 1.91 1.92 .74 1.86 1.19

W(1-r) 1.97 1.97 1.97 2.04 1.92 1.99
宽（1-r） 1.97 1.97 1.97 2.04 1.92 1.99

Number of individuals in sample
样本中的个体数

NOTES:
笔记：

b/Less than 0.005 in absoult e value
b/absoult e 值小于 0.005

e/zeros included
包括 E/Zeros

d/Includes nonworkers
d/包括非工人

e/Includes principal person.
e/包括主要人员。

TABLE A-4
表 A-4

MEAN INCOMES OF HUSBAND-WIFE AND FEMALE-HEADED
夫妻和女户主的平均收入

FAMILIES
家族

Preenrollment During-Experiment
预注册期间-实验

Experimental Control Experimental Control
实验控制 实验控制

Group Group Total Group Group Total
集团 集团合计 集团集团合计

Monthly income/100,
月收入/100，

husband-wife
夫妻关系

families:
家族：

Earnings of husband 5.76 5.50 5.63 5.81 6.06 5.91
丈夫的收入 5.76 5.50 5.63 5.81 6.06 5.91

Earnings of wife .80 .81 .81 .51 .49 .50
妻子的收入 .80 .81 .81 .51 .49 .50

Earnings of others .11 .08 .10 .15 .15
其他收益 .11 .08 .10 .15 .15

NIT payment 1.50 .08 :
NIT支付 1.50 .08 ：

Other income .12 .20 .15 .35 .39
其他收入 .12 .20 .15 .35 .39

Total family income 6.79 6.59 6.69 8.32 7.17 7.86
家庭总收入 6.79 6.59 6.69 8.32 7.17 7.86

Monthly income/100,
月收入/100，

female-headed
女头

families:
家族：

Earnings of female
女性收入

head 1.22 1.00 1.13 .73 .73 .73
扬程 1.22 1.00 1.13 .73 .73 .73

Earnings of others .06 .04 .05 .06, .09 .07
其他人的收入 .06 .04 .05 .06， .09 .07

AFDC payment 1.45 1.49 1.47 .13b/ 1.60 .72
AFDC付款 1.45 1.49 1.47 .13b/ 1.60 .72

NIT payment - - 一 .09르/ 1.65
NIT支付 - - 一 .09르/ 1.65

Other income .71 .51 .63 .68 .62
其他收入 .71 .51 .63 .68 .62

Total family income 3.44 3.04 3.28 4.17 3.19 3.79
家庭总收入 3.44 3.04 3.28 4.17 3.19 3.79

NOTES:
笔记：

e/control families received a nominal fee for participating in the experiment and submitting to interviews.
E/Control 家庭因参与实验和接受访谈而获得象征性的费用。

b/A few experimental families received AFDC, but only when a subfamily existed in the household with an eligible AFDC case. The experimentally-defined"female" head never received AFDC.
b/一些实验家庭接受了AFDC，但只有当家庭中存在一个亚家庭，并且有符合条件的AFDC病例时。实验定义的“女性”头部从未接受过AFDC。

57

TABLE A-5
表A-5

MISCELLANEOUS CHARACTERISTICS OF HUSBAND-WIFE
夫妻关系的其他特征

AND FEMALE-HEADED FAMILIES
以及以女性为户主的家庭

TABLE A-5 (continued)
表A-5（续）