Bad to worse?

In late 2022, I had a couple of posts about perceptions of moral conditions.  I concluded that assessments had become more negative in the 1960s, and after that there might have been some further decline.  It turns out that Adam Mastroianni and Daniel Gilbert were doing a similar study, which was published in Nature in 2023.  Their conclusions are completely different:

"A linear model indicated that the proportion of participants who reported moral decline was not significantly influenced by the year in which the survey was administered, b = 0.07, 95% confidence interval (CI) = [−0.11, 0.24], t(175) = 0.77, P = 0.45, adjusted R2 = −0.002, and the same model fit in a Bayesian framework indicated strong evidence of no effect (Bayes Factor of 0.04), which is to say that US Americans have been reporting moral decline at the same rate for as long as researchers have been asking them about it. (These and all tests we report are two-tailed)."

A problem with their model is that it omits a potentially important variable, or a lot of potentially important variables, depending on how you look at it--the specific question asked.*  Their sample of 177 cases includes over 70 distinct questions.  Some of these differences are small, but some are substantial.   For example, "Would you say that people are more willing, less willing, or about as willing to help their neighbors as they were twenty-five years ago?" and "In the last eight years, do you think crime has increased, decreased, or stayed about the same?" are clearly different questions, although they both involve the general topic of change in moral conditions.  So you should consider a model that includes dummy variables for the different questions in addition to the time trend.  The data set is not available to the public (the numbers are the property of the data archives), so I can't fit that model.  However, I can consider two questions that were asked frequently (data are available from the Roper Center).  One is "how satisfied are you with the direction that the country is going in at this time in terms of morals and ethics?...Very satisfied, somewhat satisfied, not very satisfied, not at all satisfied"; the other is "right now, do you think the state of moral values in this try as a whole is getting better or getting worse?"

  The figure shows positive minus negative responses to both questions over time.  For the first question, there is a clear downward trend (a t-ratio of about 10 if you regress the summary measure on time).  For the second, there's no clear trend, but the numbers are consistent with the hypothesis of a drop after 2004.  In addition to the trend, the responses for the first question show some short-term variation--e.g., opinions were more positive in October-December 2001 than in January-March 2001.  It's easy to think of an explanation for that.  

While I'm at it, here's the estimated assessment of moral conditions, adjusting for question, in the data set that I compiled.  It's not the same as the Mastroianni/Gilbert data, but there's some overlap--mine includes the "getting better or getting worse" question, but not "how satisfied are you with the direction."

*Another problem is the appeal to a "Bayesian framework" to claim that there is positive evidence of no change, but I've written about that at length here and here, so I'll leave that aside for now.

[Data from the Roper Center for Public Opinion Research]