Why Yes Men

A classic 1993 economics paper is “A Theory of `Yes Men.’“:

This paper illustrates an incentive for workers to conform to the opinion of their supervisors when firms use subjective performance evaluation. This desire to conform arises endogenously from the firm’s need to induce the worker to exert effort.

The basic model goes as follows.  A boss needs to estimate some number X (e.g., the optimal value of a design parameter).  An employee can work to get a noisy signal S about X; the harder he works, the lower his noise.  The boss also has a noisier estimate B of X, and wants to encourage employee effort to get an accurate signal S.  If no one will ever know anything more about X than these two estimates, then the boss is forced to reward the employee based on how close that employee’s estimate S is to the boss’ estimate B.  If there are several employees, the boss can also reward each employee i based on how close his estimates Si is to other employees’ estimates Sj.

Now imagine that an employee i can work to get signals not only about X, but also about the boss signal B, or about other employee signals Sj.  Under the above incentive schemes, such efforts may be richly rewarded.  Yet from the point of view of a boss who wants a good estimate of X, such efforts are a waste.  This is the problem of  (suck up, brown noser, kiss ass) “Yes Men” (and women), who pay too much attention to what the boss thinks, or to common opinion, relative to the problem at hand.

This classic paper was insightful, but I think it missed a key feature of real yes-men: most bosses of yes-men encourage, not discourage, their employees to find out what they think.  In the above model, a boss who revealed his own estimate before the employee put in his effort would have lost all ability to encourage such effort.  So such a boss should instead try hard to hide his own opinion until the last possible moment.  Similarly, if it weren’t for the benefits of teamwork, such a boss should try to prevent multiple employees from talking to each other about their estimates  Real bosses with yes-men, however, typically make their opinions clear to subordinates, and encourage full communication between subordinates.  Why?

One explanation is that bosses must signal dominance, and show they have the full cooperation of their subordinates.  Another explanation is that most bosses have other bosses above them.  And those higher bosses are likely to judge their subordinates in the same way, by comparing subordinate opinions to their own and other employee opinions.  In particular, a boss whose opinion differs greatly from his subordinates’ opinions is probably a dufus, and should be dumped.   So a boss who expects to be judged on how well he and his subordinate agree would make sure those subordinates knew his opinions.

Of course a boss would not want to be too obvious about this.  A boss who too obviously let it be known that employee opinions must conform to boss preconceived opinions, regardless of what their efforts tell them, could look and be pretty incompetent.   So a boss wants to create the appearance to his superiors that he seeks diverse opinions formed independently from his own, while actually quietly and ruthlessly squashing dissent.

But shouldn’t the boss of this boss easily be able to tell the difference between a subordinate who really seeks dissent, versus just giving lip service to such ideals?  Sure, it probably wouldn’t take that much work.  But if this higher boss must answer to still higher bosses, he might not actually want to expose such subordinate hypocrisy.   After all, if he has an ineffective dufus for a subordinate, that might reflect badly on him, who may have hired or promoted that dufus.  As long as the hypocrisy isn’t so obvious than an outsider could easily see through it, many may want to help keep up the pretense.

You might think all this would be different for the CEO, who has no boss.   But CEOs do have bosses; they work hard to keep up appearances to impress investors, customers, and suppliers.  And once investors, customers, and suppliers are affiliated with a firm, they want others to think well of that firm as well.   Prospective not-yet-affiliated investors, customers, and supplies might be more objective, except each of them usually also has a boss with opinions to consider.

It is rare to find someone with full discression and incentives to form and act on their own best estimates of what is really going on, without needing to worry about the impressions such opinions will give observers.  Those few may have the ability and incentive to see through wider yes-men hypocrisy.  But if they see it, and call the alarm, who will confirm it?   And if few confirm, who else would admit they listen to such marginalized cynicism?

GD Star Rating
loading...
Tagged as:
Trackback URL:
  • anonymous

    Robin,

    How closely do you think the above model fits the higher realms of academic economics?

  • Grant

    The signal-estimate model does seem very flawed to me, because bosses have other means of estimating how close S is to X. If the boss has a good understanding of what his employees are doing, he can look at how hard they are working, what factors they are taking into account, what alternative approaches other employees are taking, etc. In other words they can judge the process to estimate the accuracy of the end result.

    Employees are always incentivized to give their boss what they want. If the boss does not have an understanding of the employee’s work, this is going to deviate from what needs to be done.

    If the boss-employee knowledge gap is large, then it seems like prediction markets or prizes would be better mechanisms to cut down on yes-men. The boss can then reward employees based on their performance after the fact.

  • Jim Babcock

    It’s worth mentioning here that there is an established test for distinguishing thinkers from Yes Men: put out a fake opinion, and fire everyone who expresses agreement. http://www.overcomingbias.com/2008/02/my-favorite-lia.html

  • http://hanson.gmu.edu Robin Hanson

    anonymous, it fits about as well as elsewhere

    Grant, the basic problem remains even when the boss has other clues about the quality of signal S.

    Jim, yes and the point is that few bosses use such tests; they don’t want to fire yes men.

    • Grant

      True, but the model still seems incomplete without mentioning estimates of the quality of the process used to determine S.

      Is there evidence that “yes men” are a widespread problem? In the (admittedly abnormal) places I’ve worked, its seemed more fashionable to disagree with the management. Some of this behavior was likely the employee’s attempts to signal intellectual superiority over his boss (i.e., a geek always wants to look smarter than a suit). I’m thinking this is probably more common in engineering and applied sciences.

  • Pingback: Interessantes woanders (2009.06.13) › Immersion I/O

  • http://www.hopeanon.typepad.com Hopefully Anonymous

    I haven’t read the article, but from your summary it sounds like they’re folding in a bunch of possible factors that need to be teased out.

    The overall model has four categories of agents: (1) a class of subordinates (2) a class of supervisors of subordinates (3) a class of supervisors of supervisors, (4) an organization which they are all subcomponents of.

    The organization in the model seems to be the only component that has a single goal of survival. At least one of the other three categories of agents seems to have a second hardwired goal (a potential “bias”) that could conflict with their goal of survival.

    In your description and analysis of the article, you seem to provide alternate hypotheses that the second hardwired goal is rooted in either the class of subordinates*, or the class of supervisors of supervisors. For the sake of relative comprehensiveness in considering this model, it seems reasonable to also consider a hypothesis that the “bias” is rooted in the supervisor-of-subordinates (he is his own audience for his signal of dominance over subordinates).

    Also, I think the dominance-signalling model has to be teased out from the boss-is-intelligent signalling model, because those are two separable signals for traits that are considered valuable in a mid-level analyst/administrator/manager. Dominance-signalling is probably good for general administrative efficiency across a bias gradient (our primate bias against unified action in the absence of a group-accepted dominator) and boss-is-intelligent is probably valuable for more abstractly rational reasons (like “this same-priced computer produces more accurate modeling/forecasting results”.

    *Is that what you mean by this line, since you seem to distinguish it from your model of evaluations of supervisors by the supervisors-of-supervisors: “One explanation is that bosses must signal dominance, and show they have the full cooperation of their subordinates.” In other words, are you saying that the bosses must signal this dominance specifically to the subordinates, for example to ensure their full cooperation? That’s how I took it, as hypothesizing a hard-wired bias in the subordinates, as an alternative to the rest of the paragraph where you seemed to be hypothesizing a hard-wired bias in the supervisors-of-supervisors.

    Sorry about the relative incoherence and incompleteness of this comment, but real life intrudes.

  • Pingback: Overcoming Bias : Suits Show Signal Scope