1

the latter had one percent fewer increases but also one percent few

er decreases.

For variation V, a narrower interval was used for "no change."

Changes of $15/week or more were counted. This produced a substantial increase in positive changes and very little change (one percent) in the number of decreases counted. Here the experimental families had

more increases and fewer decreases--but even so, the differences do not approach statistical significance.

The result

Of the first three variations, which relate to the data used for the original report (except for correcting minor errors) variation III provides the strongest indication of greater effort, as reflected in earnings, for experimental families. Almost all families are represented, and the data have been purged of minor errors. ing differences in that table are significant at the .05 level (i.e., would happen only one time out of 20 purely by chance if there were no real difference between controls and experimental changes in earnings). Variation I was used in the original report rather than (the uncorrected version of) Variation III for two reasons. First, it involved

no assignment of values to non-response--and was "conservative" in that

sense. Second, since there were, and are, ample reasons for being

cautious in interpreting these early data, a non-significant and less marked contrast between control and experimental families, as provided in Variation I (or II), was preferred for immediate release--again with the aim of making a conservative choice.

The last two variations, which are based on a third approach to

the non-response problem and use data for full-year changes in earnings

in both cities show no significant differences. There is some effect that depends on the required size of earnings changes however. For reference it may be useful to note that a $15/week change of earnings of head and spouse corresponds to about 20 percent of average earnings at enrollment, and a $25/week change corresponds to 33 percent of average earnings at enrollment. The actual average increase over the

year was around 7 percent, broken down as shown below in Table 7.

The original report's Chart V has now been updated, thereby eliminating two small problems with the original data. There were minor entry errors in the raw data tables used to calculate the chart, and the final summation of Paterson and Passaic mean family incomes was not weighted correctly (data went to OEO for Chart V in six parts--mean incomes for each of the three incomes strata in each city--which were not properly weighted when added to get a total monthly mean income). •Neither of these errors had any appreciable impact on the Chart and the conclusions obviously remain the same. In addition to the above corrcctions, the chart has now been extended to include two additional months, bringing it to a full year. Comparisons between the original and the

updated and extended monthly means are shown below (Table 8):

Table 8: Corrected and Updated Versions of the Chart V in OEO's Original Report

The Paterson-Passaic experimental group was used alone in OEO's original chart, because their report had to be short and easy to understand and Trenton could not be "added in" in any simple or obvious way. Trenton started five months earlier and had been running longer, but there were many more observations in the Paterson-Passaic group, making the trend, though shorter, more reliable. In any case, since there are no comparable measures for the control group, Chart V can only be interpreted relative to general knowledge of income experience of the poor. Within this frame of reference, both Chart V and the comparable diagram for Trenton (or even for subgroups such as income strata) are equally emphatic in showing that there is no pronounced reduction in incomes following enrollment in our benefit program.

Further Tests and Analysis

This section presents more complete results for the subgroup of 410 families from Trenton, Paterson and Passaic from whom we have usable fourth quarterly questionnaires.

As regards the analysis of change in earnings, the discussion here will concentrate on earnings changes greater than $25/week. Similar tests and tables were compiled using the $15/week criterion but, since they generally gave the same indications (and were similarly non-significant) they are deleted here. In addition to changes in total earnings of head and spouse (called family earnings above) changes in

For instance, should one combine the same calendar month, as closely as possible, or months that are equidistant from the beginning of benefit payments?

24

earnings of the head alone have been analyzed and are presented below, again considering only changes greater than $25/week.

The earnings changes of families or heads were classified according to treatment--both the gross control/experimental contrast and distinguishing among treatments within the experimental groups (Tables 9 and 10). Chi-square contingency tests were carried out, and in no case was the null hypothesis of no difference in earnings change among groups rejected at the .10 level (a less stringent requirement than the .05 level typically used for such tests). Classifications by city, ethnic group and stratum were also made (Tables 11 and 12). The only

instance of significance at the .10 level was for husband's earnings change for the contrast between Trenton and Paterson-Passaic; here most of the difference between the two cities was found in the experi

mental groups.

When control/experimental comparisons were made within black and Spanish subgroups, there was a nearly-significant relation between the treatment and city classification and change in total family earnings. Most of this was due to a sharp difference between the two experimental subgroups (Table 13).

Vin Table 14 the control/experiment comparisons are shown within the two cities for earnings of the head alone. As will be noted, most of the favorable evidence for the experimental group comes from a disproportionate number of earnings increases in Paterson-Passaic.

Even though the other differences are not statistically signifi

cant, it will be useful to discuss further the patterns of income change shown in Tables 9-14.