Wednesday, October 20, 2010

The Ig-Nobel Prize for Economics: Should companies promote people at random?

This year, the nobel prize for economics was awarded to/shared by Peter A. Diamond of MIT, Dale T. Mortensen of Northwestern University, and Christopher A. Pissarides of the London School of Economics.  These three economists were honored for their work relating to government policies and employment and economic growth during recessions.  Among some of the many contributions in these areas are the finding that greater unemployment benefits can lead to longer periods of unemployment and the finding that obstacles to matching (in this case employers finding potential employees) are a critical factor in determining the levels of unemployment.  In fact, the research showed that problems in matching are so important to unemployment that even with extensive government spending and works programs and even in economic boom times there will always be some level of unemployment due to the difficulties of matching employees with employers.  Of course, meanwhile, the Ig-Nobel prize for economics this year went to the big Wall Street Banks for creating hard to value derivatives and credit default swaps and other financial instruments that led to the overinflated bubble that ultimately burst.  Since that really isn't research related, I am going to claim that the Ig-Nobel prize for management serve as a proxy for the prize in economics (since there is no Nobel prize for management, and business and management are related to economics so...)  This year's Ig-Nobel prize for economics/management went to Alessandro Pluchino, Andrea Rapisarda, and Cesare Garofalo of the University of Catania, Italy, for "demonstrating mathematically that organizations would become more efficient if they promoted people at random."  The premise is an interesting one, and perhaps we've all experienced this to some degree, especially if you've ever worked for a big corporation.  Companies promote managers largely based on performance (assuming you ignore any nepotism, backstabbing, or other political gamesmanship), and so the best assembly technician, data entry specialist, scientist, factory floor sweeper, etc.,  gets promoted to manager.  The problem is, being good at floor sweeping or science (or at most any other task) has absolutely nothing to do with being a good manager, and so, despite any individual person's great performance at their first job, they may be the worst manager the world has ever seen.  Believe it or not, this observation has been somewhat codified by Canadian psychologist Laurence J. Peter, and is thus named the Peter Principle, which states: "Every new member in a hierarchical organization climbs the hierarchy until he/she reaches his/her level of maximum incompetence".  If this is true, or happens somewhat regularly, it begs the question of whether or not companies should promote the best person from any given level, or, perhaps instead simply promote people at random.  In the paper by Pluchino et al., the authors tested this idea questioning whether the "common sense" method of promoting the best people (i.e. promoting those who excel most at their current level) might make a company less efficient than if it were to promote people at random.  Of course, they didn't try this in a real company, but ran computer simulations, allowing them to test the idea over and over and average out the results.  Essentially, they designed "companies" that had a pyramidal structure: lots of low level employees, slightly less middle managers, even less upper level managers, and ultimately one person who would be in charge (see figure above).  
They then had the computer software randomly generate "individuals" who had "competence" values ranging from 1 to 10, and ages ranging from 18 to 60.  If an individual was incompetent (a value less than or equal to 4) or of retiring age (60) they were removed, a spot opened at that level, and an individual from the next level down was promoted to fill the vacancy.  Several strategies were applied: 1. the "best" approach, where the most competent at a given level was promoted, 2. the "worst" approach, where the least competent person was promoted, and 3. the random approach, where the individual that was promoted was chosen at random.  Each of these strategies was applied for the two hypotheses being tested: 1. The common sense hypothesis, where an individual's level of competence transfers from one level to the next (i.e. it is assumed that good floor sweepers generally make good managers, though the authors did build in a possible swing of plus or minus 1 point allowing that some floor sweepers could be slightly worse, or even better, managers than they were sweepers.) 2. the Peter principle, where a person's competence did not transfer to the next level with their promotion, but rather competence at a new level was again randomly assigned.  Finally, the measure of success for each of these methods was a valuation of the company's "global efficiency" which was calculated by adding up the competence values at each level and weighting them more as the level approached the top of the company (basically assuming that better or worse performance at the top of the company would have more of an effect on the overall performance of the company than competence or incompetence at lower levels).  What the computer simulations showed is that when the common sense outcomes applied (that is, when competence was basically the same from one level to the next) and you promoted the best people at each level, not surprisingly, you got very good global efficiency for the company.  When the worst person was promoted, the company had pretty lousy efficiency. What was surprising was that if competence at one level had no effect on competence at another level (the Peter principle) then promoting the "best" person at each level actually resulted in the worst global efficiency, and promoting the "worst" person at each instance resulted in the best global efficiency.  Finally, under both hypotheses (common sense and Peter principle) promoting people at random resulted in small increases in global efficiency. From this, the authors conclude that, if you don't know whether common sense principles or the Peter principle is at work, your best bet would be to promote individuals at random because even though the effect was small, you would always get an increase in global efficiency rather than risk the loss in efficiency that would result from using the best strategy if the peter principle really is at work.  And, of course, since we don't know if the Peter principle really is at work, you wouldn't want to risk promoting the worst candidates only to find the common sense principle was right.  Of course, there are definitely some considerations that need to be made before instituting the random promotion policy.  First, I think the assumption that a highly competent person at one level (a 10) could be so inept at the next level to be randomly assigned a 1 and then be fired (even if the probability of this is small, since the re-assignment is not totally random, but falls along a normal distribution).  To me, if you excel at one job, you likely have skills that apply at every level (being punctual, organized, responsible, hard working, smart, easily trainable, etc.)  Therefore, I would like to see the simulations re-run with promotions in the Peter principle assigning random values between 4 and 10, rather than 1 and 10 (or at least skew the distribution more to the right).  Second, I think that even if you tweaked the game this way, and it still came out that randomly promoting people was the better strategy, one still has to consider the repercussions of a random promotion policy that might kill the incentive for workers to excel at their job (since they know it will have no impact on whether or not they get promoted).  Ultimately, I think that this would lead to the majority of employees operating at a level of competence just high enough to not get fired.  Still, the article is interesting, and suggests that the Peter principle is something that companies and other hierarchical institutions need to be wary of, and perhaps, look for a better way to assess the skills that will be needed at each new level and base promotions off of a combination of excellence at the current level and this potential for excellence at the next level.

Figures were taken from the article, the reference for which is:
Pluchino, A., Rapisarda, A., & Garofalo, C. (2010). The Peter principle revisited: A computational study Physica A: Statistical Mechanics and its Applications, 389 (3), 467-472 DOI: 10.1016/j.physa.2009.09.045


  1. Many thanks for your nice quotation of our work. I'm Andrea Rapisarda one of the authors of this study. Of course our simulation is very schematic
    and we did not include any psychological effect.
    However evolution in nature proceeds in the same way by random mutations which give a finite probability of success even to the most remote possible changes, reinforcing (and not removing) them when they are successful. We have had a lot of positive comments see here for example and we have just discovered
    also a real experiment applying a successful job rotation (very similar to what we have found). You can see here some info
    Best regards
    Andrea Rapisarda

  2. Thanks for the info, and congrats on the paper and on the award. The paper is most certainly interesting, and even more so if you are finding real life examples to support your model. I have had an intuitive understanding of the Peter principle for years, but only thought about it in the context of corporations and other hierarchical "career paths". I will have to give this some thought as to how it pertains to evolution/adaptive peaks, and perhaps how it might map to the fitness of a species based on whether or not the mutations within the population are completely random or not. In the meanwhile, perhaps I will check out the story of Ricardo Semler and his company.

  3. I am a consultant that markets talent management selection and development psychometric tools. The tools I market utilize criterion-related validation which means that are validated against on the job performance.

    Given my above context, how would you propose I use the Peter Principle in my psychometric product marketing?

  4. If you like the book Sleights of Mind I would think you might also like Dan Ariely's Predictably Irrational