Derek Hum and Wayne Simpson*

The authors are professors of economics, University of Manitoba. Derek Hum was formerly Research Director of Mincome.

Whatever happened to Canada's guaranteed income project?

Canada's transfer programs perch atop the policy agenda, especially now that universality is no longer sacred and there is interest in reforming the welfare state in favor of stronger economic incentives. This calls to mind past discussions concerning a guaranteed annual income. Canada conducted a multi-million dollar study of guaranteed income during the seventies. This experiment, dubbed Mincome, gave a number of Canadian families a guaranteed income to determine whether individuals would work less, and if so, by how much? It is natural to ask: Whatever became of Mincome and its results? Weren't we supposed to find out whether or not Canadians would work less, and if so, by how much?

This note reports Canadian findings on work effort associated with a guaranteed income. Although Mincome ended some ten years ago, these are the first full findings available. We summarize results separately for husbands, wives, and single female heads. For perspective, we compare these findings with results from the four U.S. experiments.

Mincome: how and why it happened.

Some feeling for the social experimental approach is necessary to appreciate the Mincome project. The origins of

Mincome are to be found in the United States. The negative income tax (NIT) fascinated those charged with developing a policy for the American War on Poverty and the idea of conducting an experimental test of this radical proposal rapidly gained acceptance, not only to determine the administrative feasibility of such schemes, but also the extent of any work disincentive that a guaranteed income might induce. It was widely conceded that the work disincentive question was the prime political stumbling block. Consequently, beginning in 1968, the United States embarked on a series of four large scale social experiments to explore a guaranteed income plan. At the heart of all these experiments was the extent to which a guaranteed income would cause able bodied individuals to reduce their hours of work or quit their jobs altogether.

Manitoba's original support for the GAI was grounded in its strong interest in administrative issues. Premier Edward Schreyer of Manitoba viewed the GAI as essentially involving "income-testing" and that it "didn't differ at all [from] a negative income tax". Furthermore, because the GAI "would. . . substitute for the Canada Assistance Plan Program", the Mincome "project would be established under the aegis of the Canada Assistance Plan". Premier Schreyer saw Manitoba's financial involvement at "something over $500,000" and the number of families involved "possibly 500", but "closer to 300". What emerged eventually was not the simple demonstration involving "300 families" and a half million dollars that Manitoba wanted, but an extremely complicated experiment, modeled along the lines of the American efforts and concentrating on the issue of work responses.

Unlike the American efforts however, which all eventually released final reports and findings, the Canadian project languished. The project published no official findings concerning the labour market response of participants, and the vast amounts of data collected remains archived. The Mincome project died a quiet death in 1979, officially reported as a redirection of experimental objectives. It must be remembered that Mincome's official demise came towards the end of the 70s. The social security review had ended by then; there was no political support in the country for sweeping reforms of the type promised by a guaranteed income; the GAI concept itself had lost its allure.

During the next years the fate of the data itself appeared uncertain. The manner in which the data was archived (unpublicized location, unknown means of access, etc.) was not conducive for research. Not surprising, policy debates in Canada concerning the effect of a guaranteed income on work. behaviour have had to rely on American results --- until now.

How a GAI works and the design of Mincome

A negative income tax (NIT) provides a minimum cash benefit (G) to families having no income, and reduces the payment amount by a "tax back" or benefit reduction rate (t) for each dollar of other income received. Since the family can never receive less than the amount G, this is tantamount to guaranteeing the family a minimum payment, hence the term guaranteed annual income. Advocates of a GAI have emphasized the system's lack of discretion in determining eligibility and benefits, its avoidance of stigma, its efficiency in targeting payments to the low income population, and the possibility of integrating the GAI/NIT with the positive income tax system.

Two of its potential drawbacks are its costs and effect on work incentives. Clearly, the more generous the GAI program, as given by high support levels (G) or low tax back rates (t), the larger will be the program costs. This results because non workers will receive larger payments, low-income earners will keep a larger fraction of their earnings, and a larger proportion of the population will be recipients since high guarantees and low tax back rates have the effect of raising the eligibility threshold of the program. Similarly, generous GAI benefits might induce individuals to work less, especially if the individual perceives the combination of leisure and guaranteed payment to be superior than that achieved by working. Economic theory alone, however, cannot furnish accurate estimates of the work disincentive resulting from a GAL Consequently, Canada and Manitoba conducted Mincome.

The Mincome project selected families from Manitoba (Winnipeg, Dauphin, and several small rural communities) and assigned them randomly to different GAI plans for three years. The sample took into account family structure (number of heads, one or two earners, single individuals) and normal income received (four or five levels). It also excluded families earning above a predetermined amount (approximately $13,000 in 1974 for a double headed family of four). Three support levels (G) were used: $3800, $4800, and $5800 (1975 dollars) for a family of two adults and two children. These support levels were adjusted for differing family size and structure , and increased annually. Three tax back rates (t) were used: 35%, 50%, and 75%. The (G, t) combinations of ($5800, 35%) and ($3800, 75%) were not tested. Mincome also enrolled a control group; that is, a group not eligible for payments but which provided information for research purposes. It is the presence of this control group which allows strong inferences to be drawn. For example, we can focus, first of all, on the "experimental response"; that is, did Mincome have any effect at all on the work behaviour of GAI recipients, and if so, how much? The experimental response is simply the difference in work effort between families eligible for GAI payments (treatment families) and those not eligible to receive payments (control families).

Work response: what have we learned?

Economists had a strong basis in theory for believing that a guaranteed annual income would reduce the incentive to work, which in turn would lead to higher welfare costs. This conviction was supported by a large body of non experimental evidence, primarily from the U.S., but which proved unconvincing as to the actual size of the disincentive effect that could be expected. It is one thing to argue that there is some work disincentive cost to a guaranteed income program; it is quite another matter to assert that this cost is too high without estimates of the expected response.

Indeed, the concept of work disincentive is itself loosely defined in policy discussion. Economists normally equated it with a decline in labour supply. Labour supply, however, can include time spent working, time paid for not working (holidays, illness, disability, severance, etc.), time spent looking for work, and even time spent preparing for work (education, commuting, etc.). As a practical matter, researchers have concentrated on annual hours worked as the most reliable measure. For simplicity, we shall focus solely on annual hours worked.

Broadly speaking, two research approaches have been employed, involving either non structural or structural models. Non structural models simply compare the average differences in hours worked between experimental and control groups. In theory, a random allocation of households to the various plans and the control group permits a simple comparison of means to give unbiased estimates of the experimental effects. In practice, the experiments employed an allocation scheme which gave low income families a lower chance to enroll in the generous plans. Moreover, the decision to remain in the experiment depends on the financial incentive, which in turn depends upon the generosity of the assigned plan and the income of the family so that attrition is also non random. Hence various control variables were often introduced by researchers to correct for possible bias arising from non random allocation in the experiments, but these technical refinements are considered elsewhere.

The results in Table 1 are derived from five experiments with many common features but some important differences. Each experiment, as noted, concentrated on low-income households with able-bodied heads and dependents. The New Jersey, Rural, Seattle- Denver and Mincome experiments used similar income eligibility cutoffs, about 150% of the official poverty line, while the Gary experiment admitted households with much higher incomes in order to examine truncation bias. Each experiment included multiple financial treatment plans defined in terms of guarantee levels (adjusted for family size) and tax rates. Guarantee levels range from 59% to 125% of the poverty level, and tax rates vary from 30% to 75%. Seattle-Denver included a tax rate which declined one- quarter per cent per $100 earned. Mincome enrolled 1300 families, about half of them in the control group, and was typical of the scale of the other U.S. experiments with the exception of Seattle-Denver where 4800 families were enrolled. In addition to the declining tax rate, the Seattle-Denver design included variations in the duration of the experiment and training programs. Moreover, each experiment examined poverty in a different setting; New Jersey examined inner cities in the north eastern U.S., the Rural Experiment covered rural Iowa and North Carolina, Seattle-Denver looked at the urban western U.S., Gary concentrated on black households in the urban Midwest, and Mincome looked at the urban Canadian prairies. The experiments, taken together, provide a broad overview of work responses to a guaranteed income program in a variety of settings of importance to policy makers.

Table 2. Change in Annual Hours Worked from Income Maintenance Experiments: Evidence from Structural Models.

The advantage of non structural models is their simplicity. However, this simplicity has disadvantages since the results of non structural models cannot be applied to general policy analysis. Consequently, the results in Table 1 estimate the experimental response to the particular guaranteed income plans tested, which may be different from a guaranteed income plan designed today. For example, Table 1 cannot be used to assess the Universal Income Security Proposal (UISP) proposed by the Macdonald Commission because the UISP features do not resemble the plans actually tested. Furthermore, non structural results cannot be used to determine work response to changes in the present tax-transfer system, such as for example, the 1993 Child Tax Benefit. For these purposes, structural model estimates are necessary.

A summary of the results from structural labour supply models for the U.S. experiments and Mincome is contained in Table 2. Not unexpectedly, the wage and income elasticity estimates are quite small, consistent with the small labour supply responses from non structural models presented in Table 1. Income elasticities are about 0.1 for all groups for the U.S. experiments and Mincome, (meaning that a 10% increase in income would only reduce annual hours worked by 1%) . Substitution elasticities are also quite small, never exceeding 0.4 with a mean for all groups of about 0.2 (meaning that an increase in taxes that reduces after-tax wages by 10% would decrease annual hours worked by 2% overall). Since the 10% decrease in after-tax wages would normally imply a 10% decrease in income, there would be an offsetting 1% increase in hours worked from the aforementioned income effect. Thus the total effect of a 10% increase in taxes is also a 1% reduction in annual hours worked. The important point is that the estimated labour supply effects contained in Tables 1 and 2 are quite small, certainly smaller than most studies of labour supply from non experimental behavior have suggested. Thus the preponderance of evidence from the experiments testifies for those who would argue that a guaranteed income plan, or indeed other changes in social policy which alter the tax rate and transfer income received by households, will not have large work disincentives.

Conclusion

Our purpose in this note has neither been to consider anew the merits and criticisms of a guaranteed annual income, nor to suggest that a GAI remains the best hope for restructuring Canada's current income maintenance system. That is an entirely different debate. Much is now different about Canada in the nineties compared to the seventies. Our narrow aim has been two-fold. First, we wish to record the denouement to what was unarguably a dramatic and extraordinary episode in Canadian policy history. The social experiment, Mincome, was the first of its kind in Canada, but remarkably little is known about this landmark project. Perhaps in large part, this is explained by the absence of any credible results issuing from the experiment, particularly on the question surrounding work response. Our second modest objective, then, is to report in summary fashion the first results available on work responses to a guaranteed income for Canadians, and to set these findings alongside the earlier U.S. results. Our interpretation suggests that the work disincentive effects of a GAI are quite modest. Whether such disincentives are too "large", and the costs "not worth it" depends upon the individual assessor and, ultimately, political judgment respecting public support for a guaranteed income approach. Nonetheless, that assessment should now be better informed as a result of the Mincome experience.

_______________________________

Notes: