The composition of the SIME/DIME sample was defined partly with reference to the likely population to be included in any national negative income tax program and partly to assure that important questions about the magnitude of the overall work effort response and about possible differential responses by different population groups could be satisfactorily answered.
The sample was restricted to families with heads between 18 and 58 years of age at enrollment in order to focus on those at least potentially in the prime-aged labor force. For this reason, families with disabled heads were excluded. In addition, eligibility was limited to families with total earnings of less than $9,000 a year if one head was employed, or earnings of $11,000 a year if both husband and wife were employed. These earnings cutoffs were a compromise between the need to include those, although above their breakeven income level at enrollment, could be expected to fall below it during the period of the experiment, and the wish to exclude people whose incomes were so high that potential eligibility for cash transfers appeared highly unlikely. These earnings cutoffs do raise the potential problem of sample selection bias in the results.
To be able to detect differences by family type, both one-parent families with a dependent child present and couples were included in the sample. For convenience, the latter type of family is frequently referred to as two-parent, although there was no requirement that such families contain a dependent child. Nor was there a requirement that the two family heads be legally married, although for convenience they are often referred to as husband and wives. Three ethnic groups were included in the SIME/DIME sample — Blacks, Whites, and (in Denver only) Chicanos. All families were residents of selected low-income census tracts in Seattle and Denver.
The experimental sample was designed using a sophisticated mathematical procedure to yield the maximum amount of useful information within a fixed budget. To achieve this purpose, enrolled families were divided into a large number of different types or "strata." For example, families were divided into seven different income levels, depending on their average income over a number of years prior to the start of the experiment. They were further subdivided according to race and the number of family heads present (one or two). In addition to including a large number of family types or strata, the experiments tested a variety of treatment combinations; twelve NIT treatment (the eleven NIT plans tested plus the control treatment) combined with four counseling/training subsidy treatment (three experimental plans plus the control treatment), in combination with two different periods of experimental eligibility (three and five years). In the interest of economy, the experiment did not test every possible combination of these treatments, but a large number were tested.
After determination of the treatment combinations to be tested and family types to be enrolled, the essence of the sample design problem is to determine the number of families of each family type to be assigned to each treatment combination. For a particular family type and treatment combination, this number is know as the "cell size". Obviously, the average experimental cost of families in a given cell will depend both on the characteristics of the family and the generosity of the treatment combination. Low-income families, for example, will receive higher NIT benefits under a given plan than high-income families. Families assigned to high-guarantee NIT plans or to 100 percent training subsidy plans will cost more to enroll than families in the control group or in low-guarantee and less generous training subsidy plans. It should also be mentioned that not every treatment combination tested was of equal interest or importance. For example, certain guarantee/tax-rate combinations were considered more feasible than others (basically, those in the middle range were considered the more policy-relevant), and consequently the designers put greater emphasis on obtaining precise behavioral response estimates for these combinations.
Taking account of the cost differences of the various cells and their varying degrees of policy relevance as well as their relative contributions to statistically precise measurement of the response pattern, the mathematical procedure alluded to earlier was used to determine the cell sizes in SIME/DIME. A family of a given type which had been selected for the sample was then randomly assigned to a particular treatment combination on the basis of the cell size computed by this mathematical model. Since assignment to particular treatment combinations was completely random for all families of a given family type, the hoped-for result was that measured differences in the behavior of people subject to different treatment combinations measured the effect of the treatment differences.
In all, 4,800 families were enrolled in SIME/DIME including control families. The way the initial sample was distributed by family structure and race and by assignment to site, treatment, and treatment duration can be seen in Table 3. Note that the "pure" control group (that is, the group eligible for neither the cash transfer nor the counseling/training subsidy treatments) accounts for less than one quarter of the sample because the same comparison group can be used for all the treatment variants. The groups eligible for the cash transfer treatment only and the counseling/training subsidy treatment only were slightly smaller than the control group. The group eligible for both a cash transfer and a counseling/training subsidy treatment is about twice as large as either group receiving a single type of treatment.
Table 3. The Distribution of the SIME/DIME Sample at Enrollment
Families were located through a intensive survey effort, which first identified the areas in Seattle and Denver that would be most fruitful in terms of the expected yield of eligible families and then canvassed those areas on an individual dwelling-unit basis.
During the period of the experiment, in spite of strenuous efforts to keep track of sample families and persuade them to continue in the experiment, some families dropped out. Over the first thirty months of the experiment, 20% of the originally enrolled husbands, 15% of the originally enrolled wives, and 15% of single heads of families dropped out. The husband-wife differences are due to differential drop-out rates in the cases of couples that split up.