Junk the Bell curve – but in the right places and for the right reason

Maybe there were loud cheers in the Infosys flock with the announcement that the company is doing away with the bell curve based performance assessment. And if Economic Times which ran this story is to be believed, the “decision is already showing results” with a substantial reduction in attrition. Looks like one of those Efficient Market Hypothesis scenario, where the employee attrition rate at Infosys has already factored in this “junk the bell curve” decision of the management to be implemented from this quarter. Or maybe it’s a post hoc fallacy, but that is not the issue for discussion here.

I am all for the end of Bell Curve way of thinking, but only for its universal applicability to domains where it is simply not valid leading to errors in judgment.

The normal distribution, popularly known as the bell curve, has a strong hold on our way of thinking. We almost believe it to be a law of nature (It is not). But fundamentally, it is a distribution of errors. You can see how the bell curve equation can be derived in this 5 minute YouTube video.

Frankly, Bell curve leads to a fairly benign performance appraisal system.A Bell curve is symmetric, it’s mean and median are the same - the average is truly sitting at the centre, with 50% of the values above and below it.

Where else in the world of human interactions to do you see such an egalitarian (Yes – egalitarian) distribution?

Whenever people interact, minor advantages conferred by our fickle preferences or chance snowballs to huge differences over time. This is the classic Matthew effect

Daniel Kahneman succinctly put this as his “favourite equation” –

success = talent + luck

great success = a little more talent + a lot of luck

Any appraisal system that replaces a bell curve based system, and which also wants to rank the employees for rewarding/penalising the top/bottom performers is likely to be far more discriminatory than the poor, much maligned and misapplied Bell. Employee performance, if measured by pure output, is going to show a Power Law distribution. Thomas Piketty, in his book Capital in the 21^st Century, has shown the prevalence of such skewed distribution in income and wealth of individuals.

Are organisations and performance appraisal systems willing to accept the role of luck in success, and the perception of success which leads to further opportunities for the winner in turn leading to the creation of superstars?

The graphic below shows the relative share of various segments of population for Income and Wealth in USA (for 2010, based on data from Piketty’s book), and the share under normal distribution (say of Income, with mean of USD 45,000 and standard deviation of USD 15,000).

The bell curve leads to a far lower level of inequality. 

Now if the idea behind junking the bell curve is to do away with the forced identification of the “non-performers” – then I am all for it. 

I would live in a bell curve world any day - and wish the same for the world – but fear that the Power Law is here to stay.

Call Drops in India - Does the TRAI Consultation Paper address the core issues?

Dropped calls have become a part of the telecom scene in India. We don't even get angry any more when we face a call drop- we simply accept it and call again, or if we are indoors, try calling again using the land line.
Recently, when the issue of call drops started hitting the front page with the Prime Minister of India also commenting upon the issue, the Telecom Regulatory Authority of India came out with a consultation paper on 4th September 2015 titled - "Compensation to the Consumers in the Event of Dropped Calls" . The paper sought comments from all stakeholders by 21st September 2015.
Like a good citizen and a willing stake holder (sic)  I reviewed the consultation paper, and dutifully forwarded my comments to TRAI.
Let me say upfront, I was not very satisfied with the drift of the questions - it appeared to suffer from the logical fallacy of "petitio principii"- begging the question. When you are given a finite set of options to choose from, not only is your choice constrained, but even the overall scope of options that you can imagine gets constrained. The TRAI paper is riddled with such assumptions.
Here are my responses to the questions raised by TRAI in its consultation paper.

Response to TRAI Consultation Paper No. 4/2015

Compensation to the Consumers in the Event of Dropped Calls

Q1: Do you agree that calling consumers should not be charged for a call that got dropped within five seconds? In addition, if the call gets dropped any time after five seconds, the last pulse of the call (minute/second) which got dropped, should not be charged. Please support your viewpoint with reasons along with the methodologies for implementation.

Response Q1: The calling consumer should not be charged for a call that gets dropped within a finite duration of the beginning of the call. However, the cut-off duration of 5 seconds for deciding when not to charge for the call, as suggested in the Consultation paper, is too low, and appears biased in favour of the telecom service provider. The consumer derives value from the call only when the sum and substance of the intended conversation stands completed.  What is the most appropriate cut-off duration value can be debated, but in my assessment, it should be the median call duration. An analysis of the calls made by me over last 8 months shows that less than 3 % of the calls are of less than 5 seconds duration, with the median call duration being 30 seconds. It is proposed that calls that get dropped within 30 seconds of the call initiation should not be charged.

The distribution of call duration (over last 8 months) is shown below. The X-Axis shows the call duration binned in intervals of 5 seconds, the Y-Axis shows the % of calls in each bin (as bar), and the cumulative count of calls (as line)

Q2: Do you agree that calling consumer should also be compensated for call drops by the access service providers? If yes, which of the following methods would be appropriate for compensating the consumers upon call drop:

(i) Credit of talk-time in minutes/ seconds

(ii) Credit of talk-time in monetary terms

(iii) Any other method you may like to suggest Please support your viewpoint with reasons along with the methodologies for implementation.

Response Q2: Yes, I agree with the TRAI Consultation paper that calling consumers should be compensated for the call drops by the access service provider.

While the response to Q1 addressed the issue of the dropped calls which should not be billed by the service provider, this by itself does not compensate the consumer for the poor quality of service. An additional compensation, while correcting the wrong of the inconvenience to the consumer, would also serve as a financial disincentive to the service provider. This compensation may be in the form of Credit of talk-time in minutes/ seconds. For consumers who are on a per-minute pulse, the credit could be of 1 minute, while those with per-second pulse, the credit could be of 30 seconds (being the assessed median call duration).

Q3: If the answer to the Q2 is in the affirmative, suggest conditions/limits, if any, which should be imposed upon the provision of crediting talk-time upon call drop and usage thereof.

Response Q3: As mentioned in the response to Q2, the compensation may be in the form of Credit of talk-time in minutes/ seconds. For consumers who are on a per-minute pulse, the credit could be of 1 minute, while those with per-second pulse, the credit could be of 30 seconds (being the assessed median call duration).

The only condition/limit on the talk-time credited on account of call-drop could be that the credit be consumed on the same day that it is credited.

Q4: Is there any other relevant issue which should be considered in the present consultation on the issue of call drops?

Response Q4: The following additional relevant issues should be considered in the consultation paper:

1) Quantum of penalty / financial disincentive on the TSP: Presently, a penalty of “Not exceeding Rs 50,000” can be imposed on the TSP by TRAI for not meeting the Quality of Service parameters, including call drop. This is too low to serve as a disincentive. This may be suitably increased to the point where it actually starts functioning as a disincentive.

2) Mode of computation of the Quality of Service(QoS) parameter of Call Drop: Presently, call drop is measured at an aggregate level by averaging all the call drops over the period of 1 month. The consultation paper recognises the inherent limitation of this aspect and states: “sometimes averages may not give the full picture”.  The current target of average call-drop of less than 2 % thus does not adequately ensure this quality of service for the individual subscriber, but only for subscribers on average. This parameter may be reinterpreted as quality of service for individual subscriber. TSPs may be asked to provide data on the percentage of subscribers who faced call drops in excess of 2 %, in addition to the average call drop rate. TRAI may set the Quality of service target of this parameter as: “Ensure that not more than 5 % of the subscribers face call drops in excess of 2%”. A basic statistical computation leads to a translation of this parameter as a requirement that the average call drop rate should be less than 0.9%, which in turn would lead to not more than 5 % of subscribers having call drop rate in excess of 2% (Based on an average of 200 calls per subscriber per month). This is illustrated in the normal curve shown below: