Yeah it’s fair to worry about the causal interpretation of the satisfaction results, although I think the pre-trend you mention is mostly a result of the GAM model pulling the regression line down a bit due to the later observations. Limiting the analysis to the period before, say, November 9th, shows no such pre-trend. I’m personally more worried about possible confounds such as the one we note about engagement. It could also be that the FTX crisis motivated a certain group of people to take the survey and share their dissatisfaction. So, it’s true that the design of the survey does not lend itself to easily answer questions of causality, leaving it a matter of interpretation based on the results as a whole and context.
Regarding the overall interpretation, I think it might still be fair to conclude that the FTX crisis has decreased satisfaction if we take into account the other results, including the recalled satisfaction pre and before the FTX crisis and the explicit reports of concerns related to FTX, decreased trust etc.
Yes we do (and thanks for the comment!).
First, 80.1% of respondents who were asked to answer additional questions about FTX decided to do so. This is similar to the number of respondents (83.1%) who agreed to answer the ‘extra credit’ questions prior to us adding the FTX questions. So, it does not seem like there was a large tendency for respondents to not answer the FTX questions, compared to just a general tendency to not answer extra questions at all.
Second, we looked at whether there are differences in demographics between those who answered the FTX questions and extra credit questions (prior to FTX). We found that men were slightly less likely to answer the extra credit questions (81.8% for men and 85.7% for non-men) but more likely to answer the FTX questions than non-men (81.9% for men and 76.3% for non-men).
We also find that more engaged respondents are more likely to answer the additional questions (no surprise there).
Looking at cause prioritization, we found that higher ratings of longtermist causes and higher ratings of neartermist causes were associated with being less likely to answer this section. So, these do not seem to show a distinctive effect of longtermism or neartermism. This could simply be explained by a third factor or reflect an association with general ‘cause positivity’ (e.g. people who generally tend to rate all causes more highly are less likely to answer these questions).
We conducted a further analysis based on a cluster analysis which we had conducted previously (and which we will outline in the forthcoming cause priortization post). This groups respondents into three discrete groups: longtermists, neartermists, and a group which rates both cause areas highly (this group is disproportionately newer EAs). In this analysis there was no significant difference in likelihood to answer these questions, but people in the ‘high everything’ cluster were substantially less likely to answer these questions.
We did not find any statistically significant differences regarding race (White vs. Non-white) and age.