I don’t think that really explains the observed pattern that well.
I agree that in general, people not appearing in the EA Survey could be explained either by them dropping out of EA or them just not taking the EA Survey. But in this case, what we want to explain is the appearance of a disproportionate number of people who took the EA Survey in 2018, not taking the EA Survey in 2019, among the most recent cohorts of EAs who took the EA Survey in 2018 (2015-2017) compared to earlier cohorts (who have been in EA longer).
The explanation that this is due to EAs disproportionately drop out during their first 3 years seems to make straightforward intuitive sense.
The explanation the people who took the EA Survey in 2018 and joined within 2015-2017 specifically, were disproportionately less likely to take the EA Survey in 2019 seems less straightforward. Presumably the thought is that these people might have taken the EA Survey once, realised it was too long or something, and decided to not take it in 2019, whereas people who joined in earlier years have already taken the EA Survey and so are less likely to drop out of taking it, if they haven’t already done so? I don’t think that fits the data particularly well. Respondents from the 2015 cohort, would have had opportunities to take the survey at least 3 times, including 2018, before stopping in 2019, so it’s hard to see why they specifically would be less likely to stop taking the EA Survey in 2019 compared to earlier EAs. Conversely EAs from before 2015 all the way back to 2009 or earlier, had at most 1 extra opportunity to be exposed to the EA Survey (we started in 2014), so it’s hard to see why these EAs would be less likely to stop taking the EA Survey in 2019 having taken it in 2018.
In general, I expect the observation may have more than one explanation, including just random noise, but I think higher rates of dropout among particular more recent cohorts makes sense as an explanation, whereas these people specifically being more likely to take the EA Survey in 2018 and not in 2019 doesn’t really.
I don’t think that really explains the observed pattern that well.
I agree that in general, people not appearing in the EA Survey could be explained either by them dropping out of EA or them just not taking the EA Survey. But in this case, what we want to explain is the appearance of a disproportionate number of people who took the EA Survey in 2018, not taking the EA Survey in 2019, among the most recent cohorts of EAs who took the EA Survey in 2018 (2015-2017) compared to earlier cohorts (who have been in EA longer).
The explanation that this is due to EAs disproportionately drop out during their first 3 years seems to make straightforward intuitive sense.
The explanation the people who took the EA Survey in 2018 and joined within 2015-2017 specifically, were disproportionately less likely to take the EA Survey in 2019 seems less straightforward. Presumably the thought is that these people might have taken the EA Survey once, realised it was too long or something, and decided to not take it in 2019, whereas people who joined in earlier years have already taken the EA Survey and so are less likely to drop out of taking it, if they haven’t already done so? I don’t think that fits the data particularly well. Respondents from the 2015 cohort, would have had opportunities to take the survey at least 3 times, including 2018, before stopping in 2019, so it’s hard to see why they specifically would be less likely to stop taking the EA Survey in 2019 compared to earlier EAs. Conversely EAs from before 2015 all the way back to 2009 or earlier, had at most 1 extra opportunity to be exposed to the EA Survey (we started in 2014), so it’s hard to see why these EAs would be less likely to stop taking the EA Survey in 2019 having taken it in 2018.
In general, I expect the observation may have more than one explanation, including just random noise, but I think higher rates of dropout among particular more recent cohorts makes sense as an explanation, whereas these people specifically being more likely to take the EA Survey in 2018 and not in 2019 doesn’t really.