Thank you for your comment!
Indeed, we did take the average of the logs instead of the log of the averages. This doesn’t change the end and start point, so it wouldn’t change the overall decay rate we estimate. We could do more complex modelling where effects between KLPS2 and KLPS3 see small growth and KLPS3 and KLPS4 see large decay. I think this shows that the overall results are sensitive to how we model effect across time.
See Figure 4 of the appendix, which shows, whether in earnings or in consumption, that the relative gains, as shown by the log difference, decrease over time.
We used the pooled data because it is what GiveWell does. In the appendix we note that the consumption and earnings data look different. So, perhaps a more principle way would be to look at the decay within earnings and within consumption. The decay within earnings (84%) and the decay within consumption (81%) are both stronger (i.e., would lead to smaller effects) than the 88% pooled decay.
Thank you for sharing this and those links. It would be useful to build a quantitative and qualitative summary of how and when early interventions in childhood lead to long-term gains. You can have a positive effect later in life and still have decay (or growth, or constant, or a mix). In our case, we are particularly interested in terms of subjective wellbeing rather than income alone.
I am a bit rusty on Bayesian model comparison, but—translating from my frequentist knowledge—I think the question isn’t so much whether the model is simpler or not, but how much error adding a parameter reduce? Decay probably seems to fit the data better.