Get back

 A couple of weeks ago, the New York Times published an op-ed by Adam Grant, Marissa Shandell, and Courtney Elliott called "The Secret Reason Bosses Want Everyone Back in the Office, Every Day of the Week."  A few days ago, the Times promoted it in an e-mail, saying "Here’s what six years of data says is actually driving the push to get employees back to the office."  The reason they propose is narcissism--that many CEOs are narcissists, and narcissists want people to display deference to them and find displays of deference more gratifying when they are done in person.  The op-ed is based on a paper by Shandell, Elliot and Grant published in Organizational Behavior and Human Decision Processes with the title "Worship Me at the Office Altar."  They collected statements about remote work and several qualities that they regarded as measures of narcissism for a sample of about 250 CEOs of Fortune 500 companies and found that CEOs who ranked higher in narcissism were more negative in their comments about remote work.

Their argument for a connection seems unconvincing to me.  Fortune 500 companies are large organizations, so the CEO never has face-to-face contact with the average employee.  If the general claim about narcissists is correct, it would be relevant to the top management team--CEOs who rank higher in narcissism would want them to return to in-person work more quickly--but wouldn't have any implications for ordinary workers.  However, they do appear to have strong evidence of a connection (t-ratio of about 4), so how do I explain that?  

One factor is that the usual t-ratios assume that the errors are normally distributed.  In fact, the residuals from the regression aren't normally distributed:  

With an error distribution like this, the standard errors will tend to be underestimated so the t-ratios will be overestimated.  There is no perfect way to address this, but I tried a few reasonable alternatives and they produced t-ratios of about 2.5.  That's less impressive, but still reasonably strong evidence by conventional standards.  

The more important factor involves their measure of narcissism.   They propose four indicators:  size of signature on the annual reports, size of the CEO photo, cash compensation relative to the second-highest paid executive, and non-cash compensation relative to the second-highest paid executive.  If you have multiple measures of the same underlying concept, all of the measures should be positively correlated with each other.  In fact, the correlations among these variables range from -.113 to .160, with an average of .044.*  There are two possible interpretations:  either there is no underlying quality to measure, or that at least three of the four "indicators" are not good measures of that quality.  I'll go with the poor measures interpretation:  in particular, even CEOs don't have complete control over their own compensation or the compensation of other managers.   

 If you include the individual variables as predictors in the regression, relative cash compensation consistently is statistically significant, relative non-cash compensation sometimes is, photo size sometimes scrapes in as "significant at the 10% level," and signature size is never close.  So there's pretty good evidence that CEOs who are highly compensated (relative to the next highest paid employee) expressed more opposition to remote work.  How would I explain this?  My proposal is that managements generally wanted to get employees back to in-person work--maybe not all the way back, but faster and farther than the typical employee would want--and that more prominent CEOs took the lead in delivering the bad news.  "Prominence" could be understood as status within the profession--high compensation is a sign that someone is regarded as an outstanding leader.  Or it could be within the organization:  high compensation could indicate a hierarchical organization in which nothing was final until you heard it from the big guy.  In less hierarchical organizations, the CEO could stand back and let subordinates deliver the bad news.   

Although I disagree with their interpretation, Shandell, Elliott, and Grant deserve credit for collecting and compiling the data (the link is on p. 5 of their paper, which I think is open access).  Unfortunately, they don't include the names of the companies, which would let other researchers add to the data set.  Confidentiality isn't an issue, since all of the data are from public records.  

*If you're interested in Cronbach's alpha, it's about .125.