4.3
Robustness Analysis
We found some mean differences between the S-treated and S-controls in section 3. The
S-treated consists of more women, they are younger, and they had more historic sickness
absence than the S-controls. Since receiving DB is often foregone by long-term sickness,
the effect of being prioritized to Sassam on DB could be biased upward if these
differences are not controlled for. From the analysis we could however not find any
difference in results when we did not add control variables.
One concern still is that there is some unobserved difference stemming from the initial
screening that also is correlated with receiving DB. To test for this possibility we repeat
the analysis of the effect on DB by making use of the individuals flowing into sickness
benefits (i.e., longer than 14 days of absence) between 1 September and 30 November
2007. The total population includes all “potentially treated” and consists of 87,136
individuals.
The total population and the experiment populations are very similar with regard to
gender, age, and educational level (see Table A1 in appendix A for the total population).
As expected, the experimental population has more historic sickness absence. The total
population is more balanced than the experimental population. However, women are still
overrepresented among the S-treated (see Table A1). The reason for this must simply
just be accidental.
The results from estimation of the logit regression model (1) are displayed in Table 5.
Since only about 15 percent of the total population was included in the experiment, the
effects of being prioritized should be considerably attenuated. This result is also confirmed
in Table 5.
16
Although smaller, the effect is positive and significant from September and
onward. The results thus strengthen the results obtained in the analysis in section 4.2. In
order to get an understanding of these quite counterintuitive results, that is, an increased
likelihood of receiving DB when being prioritized to Sassam, we set up a theoretical model
in the next section, which is empirically tested in section 6.
16
We have also repeated the analysis with the other outcomes and with AM, but we did not find any
significant results.
