My short answer is: you can’t. At least not using a bell curve (a.k.a. forced-rating system). Ajit Vidyadharan posted this question in the Organizational Development Question and Answers section of LinkedIn:

A genuine top performer has to be rated lower to fit in the “Bell Curve” and at the same time an average employee in another sub-group is rated higher because of this same “Bell Curve”; how could you do justice to the top performer? Both these employees belong to the same vertical but separate sub-groups. And the normalization is done in sub-groups. (Ideally it should be done vertical / business-wise. But this is not an ideal situation).

My long answer is in two parts: usage and statistics

USAGE:
The ‘bell curve’ is simply a math model that consistently represents the variation in measurements of similar things in the world. What’s the distribution of student test scores in a classroom? What’s the distribution of ages in a town? When grouped along a horizontal axis, the distribution of test scores and ages form a bell curve. This property of numbers happens so often it’s called the ‘normal distribution’. But there’s a HUGE difference between testing the distribution of scores and ages to learn if they form a bell curve versus instead assigning scores and ages to fit a bell curve. Just because many things in the world form a bell curve doesn’t mean that all things do. And when you force a model on reality, well, reality has a funny way of not cooperating. Ergo the scenario you’ve described in your original question. I’d suggest that any organization that insists on squeezing it’s employees into a forced-rating or bell curve model has missed an opportunity to use a more meaningful and congruent way to determine employee performance.

STATISTICS:
Like any statistical test or tool, the normal distribution requires two things to be meaningful: (1) a sufficient sample size and (2) commonality among the things objectively counted/measured. In the scenario above, the distribution of test scores in a classroom of only 10 students won’t likely form a bell curve, but the distribution of test scores in a classroom of 50 students may. To say with meaningful confidence that the distribution of test scores in a classroom forms a bell curve, you’d need at least 100 students. Similarly put, to test if employee performance mirrors a bell curve, you’d need to evaluate some common objective measure of performance for at least 100 employees.

As for what constitutes a ‘common objective measure of performance’ that’s a whole other can of worms given the diversity of our work force and the innumerable other unmeasured factors that can affect an employee’s perceived impact (their manager, organizational issues, non-work distractors, etc).

The paradox of using the bell curve to rate employee performance is that it has the appearance of objectivity, but misuse and a poor grasp of statistics often reduces the exercise to a totally subjective one. If you have to fit me into a curve with my peers, and you don’t have a common measure or sufficient group size to do so objectively, how are you going to do it?

This HBR article offers some more ideas about the effectiveness (or lack thereof) of forced-rating systems. more below for more ideas.

And if you’ve seen forced-rating systems used well, we’d love to hear about your experiences in the comments section below.