This is a story of a Human Resources head at an IT products firm who had joined the company in April.
“April is appraisal time, and perhaps a wrong time to enter any company, especially if you are in HR. As soon as I started work, I was expected to finish appraisals, summarise them, and recommend pay hikes and promotions for around 200 people. Things were two months behind schedule. I told the respective bosses to use the existing appraisal forms and finish the process, as it was too late to implement the new appraisal system that I wanted. That would have to wait till the following year.
One day, the Business Development head peeped in.
“Hi!” he said, “You got our appraisals, right? You know, all our 12 guys have done well. We received major orders. Just one guy is somewhat behind the others, but he supported others well.”
“Good to know”, I replied, “ What’s your point?”
“You know that ‘bell curve’, where I am supposed to put some 5 per cent people on top grade and 5 per cent in bottom grade and rest in the middle. I couldn’t do it. It’s not fair.”
I wondered if there was any logic in forcing a bell curve on a small set of twelve people.
Shortly afterwards, the Product Development head dropped by, and had the same issue to discuss. He said, “I have gone by the bell curve and rated most of our people as average or above average. But it was tough to explain this to them. They seem to have lost interest. The good ones already have their CVs in circulation, anyway. They know they can’t be in the A-grade every year, even if they do well, because the slots have to go to others by rotation.”
With this, I knew I had a bigger problem on hand: attrition. A ‘bad attrition’ of good people leaving the company and the mediocre hanging around.
To discuss this I met the company’s CEO. The CEO said: “Our appraisal system, based on the bell curve, has been in place for over three years. Some large and reputed companies also use it. Why should we change it?”
I knew such companies. Their success justified all their ‘best practices’. That year, I couldn’t convince the company to do away with the bell curve. After an exodus of talented employees, the company was forced to junk it. When some high performers left, even those who were fitted in good grades followed. It is not unusual for such people to stick together because they have confidence in one another’s abilities.”
Theory vs practice
The ‘theory’ is that if you map people’s performances, most people will fall into the middle band, which comprises average performers. According to the Bell Curve theory, the number of people you find in each band away from the average will keep reducing as you move away from it. This is supposed to happen in both directions. The shape you thus get is of a bell. The bell curve represents the number of people in each performance band or grade.
Researchers found that individual performances in their samples had a good fit within a bell curve (also known as Gaussian distribution, or normal distribution).
Curse of statistics
Today, the bell curve, as a method of grading people’s performance, stands discredited. Many companies have distanced themselves from this system. This should have been obvious then. The proponents of the theory failed to see that it was only fit for some data sets, and that the performances of all sets of employees would not fit the bell curve. Call it the ‘Curse of Statistics’.
The bell curve is just one of the many patterns that statisticians use to describe how variations of a property, a trait or parameter, related to things or people, may look when drawn as a graph. The height of individuals in a given set of people, temperature at a particular time of day at given place in a particular season, happiness index or consumer confidence index of a group of people are some examples.
One of the jobs of statisticians is to find out what type of distribution best describes observed variations of a parameter. This is to find the ‘best fit’. The statistician is trying to find what model matches best with the distribution of a given data set.
Past and future
In other words, the model will represent what exists or existed and this does not automatically mean it is predictive of how a future set of data will look.
Unless cause and effect relationships have been identified, the distribution should not be used to make decisions about individuals or individual instances.
Big DataThe limitations of the bell curve (or any other distribution) should be clear. These are:
1. Measuring people’s performance is a highly subjective and complex exercise. When we attach a number to it we brush subjectivity under the carpet.
2. The factors that affect performance are many (attitude, skill, opportunity, culture, processes, motivation, and so on) and their measurements are subjective. These might not be cause and effect relations; we can expect them to be changing, non-linear, and even circular (effect becoming a cause).
3. Comparing performances across functions like sales, development, customer support, and so on, is also subjective and complex.
Some researchers recommend a power law pattern of distribution: a small number of people contribute abnormally more to a firm’s success (something like Pareto principle). But even this model can’t be applied universally.
It is possible that employee performances in some companies at some point of time will match the ‘bell curve’ distribution. However, there is no certainty that the bell curve will remain always remain the ‘best fit’.
Consider the possibility: the very act of using bell curve to decide on grading and pay hikes, and so on can change the distribution of performances. High-performing people may leave since some of them won’t be put in the higher bands. Average performers may not be motivated to improve their performance. The average performance level itself may fall. Such things can get masked if a company is in a growing market and has built a momentum of good market shares, and may not be recognised until it is too late.
Measuring employee performance and using it for salary and promotion needs ‘systems’ in place to avoid basic errors. But, above all, it needs sensitivity and human judgment. Statistical tools can’t substitute judgment. The HR head in the story recounted earlier was right in being wary about the bell curve and questioning its usefulness as a one-size-fits-all model.