I love getting questions: It’s a great way to find out what’s on people’s minds so I can cover the issue that’s been raised in my writings on performance management or in my Making Performance Appraisal Work seminar.

This morning, the “Contact Us” link at www.GroteConsulting.com delivered a polite question that asked why I argued in favor of something that in fact I don’t believe in at all: a normal distribution of performance appraisal ratings. Here’s precisely how Rajiv from the Indian Institute of Management put it:

Mr. Grote,

I am curious about this argument you have put forth. My question is: what is the basis for you to think that employee performance in companies follows a normal distribution?

I’d be grateful for your views on my question.

Regards,

You know what a “normal distribution” is. It’s the classic bell-shaped curve that applies to almost everything in life. I think mathematicians call it a “Gaussian curve.” Applied to performance appraisal, a normal distribution means that most employees will get whatever the form’s middle rating is (maybe “Fully Successful” or “Meets Expectations.”) A significantly smaller number will get the rating just higher than the middle (“Superior” or “Exceeds Expectations”). And exactly the same smaller number will get the rating just lower than the middle (“Needs Improvement” or “Meets Some Expectations.”) Finally, a tiny number of employees will get the absolute top rating (like “Distinguished” or “Greatly Exceeds Expectations”) while the exact same tiny number will get the lowest rating the form offers (perhaps “Unacceptable” or “Fails to Meet Expectations”). Convert those numbers to a graph and you’ve got your classic normal distribution curve.

But that normal distribution curve in bogus in organizations.

Here’s how I responded to Rajiv’s question:

Greetings, Rajiv!

Thanks for raising the question about distributions.

But I’m puzzled. I don’t think I have ever argued that the distributions of performance appraisal ratings should follow a normal, bell-shaped, Gaussian distribution curve.

As you know, for a normal distribution curve to be valid, there must be two pre-conditions. First, there must be a sufficiently large population. In most companies there are enough employees for that condition to be met.

The other condition is that there be random distribution. But in companies, you don’t have random distribution. Companies don’t hire people at random (for example, every 14th applicant) or promote people at random (for example, on an alphabetical basis). And companies provide training and coaching to help people improve their performance.

Therefore, in a well-managed company with good supervisors and tough performance standards, you should NOT expect a random, bell-shaped curve distribution of appraisal ratings. You should expect a slight positive skew.

Here’s how I explained it in my book, Forced Ranking: Making Performance Management Work (Harvard Business School Press, 2005, page 148):

Allow some flexibility in the distribution expectations. In GroteApproach®, the web-based performance management system my company has developed, clients can determine whether they want to show a recommended distribution and what that distribution should be. But here’s our default recommendation:

Distinguished — Up to 10%

Superior — About 20 – 30%

Fully Successful — About 60% or more

Needs Improvement — About 10 – 15%

Unsatisfactory — Less than 5%

This set of guidelines provides for twice as many people to be rated in the Distinguished and Superior category than in the Needs Improvement and Unsatisfactory categories. It provides a range and not a fixed requirement at every rating level. It provides for a reasonably normal distribution but one with an appropriate positive skew. And finally, as the heading indicates, the distributions displayed are not rigid requirements but instead represent the “likely percentage of employees” that will end up in each area.

I hope this helps. Thanks for writing.

I sent my email off to Rajiv. I don’t know whether I’ll be successful in convincing him that you shouldn’t expect a normal distribution of appraisal ratings. But if the system’s working right, there should be around twice as many people getting higher ratings than lower ones, with most employees in the middle.

First objection—it’s arbitrary. Well certainly using a predetermined distribution (like top 20 percent, vital 70 percent, and bottom 10 percent) is arbitrary—and that’s its great value. Using fixed and arbitrary percentages forces managers to make tough decisions about who’s an A player, who’s not, and why not. Otherwise, as happens in too many performance-appraisal systems, everyone gets rated superior, managers never have to have tough conversations about performance, and the organization slowly slouches toward mediocrity. Restricting the number that can fall into the A category, and demanding that managers identify a bottom 10 percent who, relative to their peers, are weaker performers, ensures that top talent is recognized and that those bringing up the rear have no false sense of security.

Of course, if you ranked a hundred people using a 20-70-10 process, #21 would be much closer in performance to #20 than she is to #90. That’s why companies that use the forced-ranking process tailor the actions they take with individuals to the individuals themselves, not just to which ranking bucket the person ended up. When I write scripts for managers to use in letting people know how they came out in their company’s A, B, and C player analysis, I develop five scripts, not just three: for the solid A player, the B+ (the #21 guy and his counterparts), the genuine B, the B- (the ones who barely avoided falling into the C category), and finally the true C level performer.

But it is important to use buckets in making relative comparisons (e.g., top 20, vital 70, bottom 10; or quartiling, or some similar scheme). Never ask managers to precisely rank their people in exact performance order. It’s impossible to distinguish between #20 and #21, and the totem-pole approach (who’s #1, who’s #2, and so on down until the last and worst performer is fingered) generates highly valid concerns about accuracy.

Yes, forced ranking is an imperfect process, as is any process in which fallible human beings must make tough decisions in an arena where solid, unarguable, quantitative data don’t exist. The forced-ranking process requires the exercise of honed, objective managerial judgment in a situation where information is always incomplete and the facts are sometimes contradictory. But managers make decisions based on limited data all the time—which projects to fund, which to shelve; when to react swiftly to a competitor’s move, when to let time take its course. Just because a decision isn’t based on countable units doesn’t mean it isn’t objective. Employee ranking is not the same as solving an algebra problem—it can’t be reduced to a mathematical formula.

About the Author

Dick Grote is one of America’s most successful and best-known authors, consultants, and business keynote speakers on performance management. He is the Chairman and CEO of Grote Consulting Corporation.