used to estimate the impact of the program. To estimate the impact, we compared the average outcomes of interest, such as attendance between the two groups. Recall, due to random assignment, any difference in outcomes between the two groups can be attributed to the program, that is, the introduction of the biometric tracking system. Here is what we found. First, we found a small increase in the attendance of medical staff. On the y-axis in this graph, you see the percentage of health staff present on average during our random checks. This graph includes doctors, nurses, lab techs, and pharmacists. The blue bar shows that the average attendance of the health staff in the control group was 37%. And the orange bar shows that the attendance of health staff in the treatment group during our study was about 40%. So there was a small 3% increase in attendance of medical staff in the treatment versus comparison, also known as the control group. But the difference is very small. And the overall level remained much below what the government had expected would be the impact of this program. Digging a little bit deeper, we find that this small increase is due to nurses, lab techs, and pharmacists. See the left graph, where attendance was 45% versus 40%. But if you now look at the right-hand-side graph, you see that, disappointingly, we do not find any effect on the attendance of doctors. In fact, doctor attendance was 29% in the treatment group versus 31% in the comparison group, that is, doctors in the treatment areas were not more likely to be present than in the comparison group. Looking at data from different followups, or random checks, separately, we also see that the effect on attendance can be seen in the fifth and the sixth followup but went away after that. So it seems that even the small effect on all health care workers, on average, fades away over time. The government was fully ready to scale this program up throughout the 2,200-plus primary health centers across the state, including a massive rollout of trainings and motivation camps. Yet, since the impact evaluation showed that the program did not improve doctor attendance, the government decided to, in fact, scale down the program. This saved at least $1 million US dollars of taxpayer money in just the first year and in just the fixed costs, compared to a total evaluation cost of about $200,000. Kudos again to the government for piloting this program and evaluating it, rather than scaling it out in one shot. We just went through one example of how results from an impact evaluation inform the government to scale down a program. But this is, in fact, not the usual story you will hear, and that is why I thought I would start here. And that is why I thought this might be an interesting example for you. There are, in fact, numerous examples in the opposite direction, of how programs that have been evaluated and found to be impactful have, in fact, been scaled up, including programs that were originally evaluated in one context, that have been adapted and scaled in other contexts. And even that is just one pathway from evidence that comes from impact evaluations to policy change. I will not go into detail here but refer you to our website linked in the lecture notes for other pathways.