In GDP we trust

 A paper published in the Lancet recently considered a large number of variables and found that only a few of them helped to predict national differences in Covid rates.  One of those was interpersonal trust, as measured by a question asking which statement you agree with:   "most people can be trusted" or "you can't be too careful in dealing with people."  This is a plausible finding, since the average level of trust, as measured by this question, is correlated with lots of good things at the national level.  Ezra Klein, discussing this study in the New York Times, suggests that low levels of trust explain why the United States has had higher Covid rates than many other nations:  "Now that we know the truth about ourselves, and the havoc our divisions will wreak on any pandemic response, the problem we need to solve becomes clearer.  What does good pandemic policy look like for a low-trust, high-dysfunction society?"  

My initial intention was just to point out that the United States isn't really a low-trust society by international standards--it ranks 17th out of the 80 nations for which data are available in the source used by the paper (the World Values Survey).*  But it occurred to me that trust tends to increase with GDP, so I should consider the level of trust relative to GDP. Here is a plot of the relationship:

 The United States is a little below the level of trust that one would expect from our GDP, but not really unusual.  There are some economically developed democracies with lower levels of trust than the US--I indicate three, France, Italy, and Taiwan.  I also indicate three affluent nations with relatively high levels of trust--Finland, Denmark, and Norway.  Sweden and Iceland are also high, so there's a clear pattern here.  I also mark two real outliers--China, with much higher levels of trust than predicted from GDP, and Singapore, with much lower levels.  

The Lancet paper began by regressing rates on "some key known drivers of infection," one of which was GDP.  The next stage used the "adjusted" values--basically the residuals from the regression of infection rates on the "known drivers"--as dependent variables and regressed them on trust.  Of course, this isn't how it should be done--you should regress infection rates on trust and GDP together, so I did that analysis.*

The estimated effects of trust (and per-capita GDP) on infection rates:

                   (1)                  (2)                 (3)             (4)

 GDP                              -.0068           -.0080        -.0091

                                        (.0017)        (.0018)        (.0021)

 Trust         -7.94            -4.68              -3.15          -2.31

                   (1.55)           (1.85)             (1.99)        (2.18)

The first column is a regression of adjusted infection rates on trust.  The estimates I get seem to match those in the Lancet paper (I can't say for sure, since they showed them in a figure rather than giving the numbers).  That way, it seems that trust definitely makes a difference (a t-ratio of about 5).  The second is a regression of the unadjusted infection rates on GDP and trust.  The estimated effect of trust is smaller and the t-ratio is about 2.5.   It seems that the Lancet paper didn't adequately control for GDP--I believe that they used the log of GDP, and the relationship is closer to linear.   

The third column is the same model as column 2, but omits China, which reports very low levels of Covid and high levels of trust.  Some observers have already questioned their Covid figures.  As far as trust, there's a question of whether people believe that their answers will really remain secret.  The Chinese authorities don't have much tolerance for dissent, and seem quite efficient in keeping track of people.  Thus, I suspect that many respondents are giving the answers that they think they "should" give.   Without China, the estimated effect of trust is smaller and not statistically significant at conventional levels.  

Finally, column 4 omits China and also Singapore, where the GDP per capita is much higher than in any other nation ($98,000--Switzerland is second with $71,000).  Although Singapore is certainly an affluent nation, the GDP figures for small nations can be distorted.  Also, the relationship between GDP and infection has to be non-linear (otherwise there will be negative predicted values) but given the gap between Singapore all other nations in the data we can't really estimate the nature of the non-linearity.  Now the estimate for trust is even smaller and the t-ratio is just a bit over one.

So although it's plausible that trust makes a difference, the evidence isn't very strong.  What clearly does make a difference is general affluence, as seen in this figure. 

 

 

 Returning to Klein's issue, the United States has had higher rates of Covid than expected given our economic level, and we have certainly been a "high dysfunction" society in terms of the politics of the issue.  The question is whether those politics reflect a general social problem of low trust (or low solidarity).  I don't think they do.

 

*The paper didn't give any information about the distribution of trust (or other predictor variables), so it's understandable that Klein didn't realize this.

**Ideally, I would have included the other "known drivers," but the paper didn't provide those data.  It gave Covid rates, and I got trust from the WVS (their source) and GDP from the World Bank (they weren't clear on what their source was).