ISS boards regularly ask candidates to explain statistical ideas in plain language: a p value to a layman, the difference between correlation and causation, why samples work at all. The transcripts prove the pattern: real candidates were asked for practical life examples of the normal distribution, everyday uses of the harmonic mean, which average best describes income in India, and whether a Type 1 or Type 2 error is more serious.
The logic of the genre is beautiful and slightly ruthless. Anyone can recite a definition. Only someone who understands an idea can say it simply. Day 12 trains exactly that muscle, on the concepts boards reach for most.
The method before the answers
One rule produces every good simple explanation: one everyday situation, one sentence of meaning, and stop. No formula, no jargon, no second example unless asked. Practise each answer below aloud, because the difference between knowing and saying is the entire test.
P value, to a layman
Try this shape. Suppose I claim a coin is fair, and it lands heads nine times in ten. The p value answers one question: if the coin really were fair, how surprising is what I just saw? A tiny p value means very surprising, so I start doubting the claim. It measures surprise under an assumption, nothing more.
And because boards probe, hold the two honest caveats in reserve: a p value is not the probability the claim is true, and crossing a threshold like five percent is a convention, not a law of nature. Offering a caveat unasked, in one sentence, is how a candidate signals depth without showing off.
Correlation versus causation, with an Indian example
Two things moving together is correlation. One thing driving the other is causation, and the gap between them is where careless policy is born. The example that lands: districts with more ice cream sales may report more heat strokes, but ice cream causes nothing; summer drives both. In data language, a lurking third variable can manufacture a relationship. A statistician’s job, and an ISS officer’s job, is to refuse the jump from moved together to caused, until design or evidence earns it.
Which average, and which error
The income question a real board asked has a clean answer: the median. Income in India is highly skewed, a few very large values drag the mean upward, so the mean describes almost nobody while the median describes the middle Indian. Average is a choice, and the choice is the answer.
The error question, asked in the 2024 cycle, rewards an example over a definition. A Type 1 error convicts an innocent claim, a Type 2 error acquits a guilty one. Which is worse depends on the cost: in cancer screening, missing a real disease, the Type 2, is usually the expensive one; in criminal courts, punishing the innocent, the Type 1, is the one the system fears most. Saying it depends, and then showing exactly what it depends on, is a full marks answer.
Sampling, normal curves and the mental arithmetic
Why does asking a few thousand households tell us about a billion people? Because a well drawn sample is a fair miniature of the whole, the way one spoon tastes a whole pot of dal, provided the pot was stirred. The stirring is randomisation, and sampling error is the honest admission that a spoon is not the pot, which is why estimates carry margins.
The normal distribution’s everyday example, asked in a real interview: heights of adults cluster around an average with symmetric tails, and so do many quantities built from many small independent pushes, which is the intuition behind why the bell curve appears everywhere and why averages of samples behave so predictably.
And one final verified warning from the transcripts: small mental calculations are real. A candidate was asked the harmonic mean of 1 and 2 without pen and paper. The answer is four thirds, and the deeper preparation is composure: slow and correct beats fast and wrong, in that room and in this service.
Tomorrow, Day 13: the year’s data current affairs. Census 2027, the DPDP rules and AI, in interview ready form.
For the free ISS Interview Starter Kit and the DAF Question Mapper, send the word INTERVIEW on WhatsApp to +91 842591 5355.
Frequently Asked Questions
How do you explain a p value in simple words?
It measures surprise: assuming a claim is true, how unlikely is the data we actually saw? A very small p value means the data is very surprising under the claim, which weakens the claim.
What is a simple example of correlation versus causation?
Ice cream sales and heat strokes rise together, but neither causes the other; summer drives both. Correlation is joint movement, causation is one thing driving another.
Which is more serious, a Type 1 or Type 2 error?
It depends on costs. In disease screening, missing a real case, a Type 2 error, is usually worse. In a courtroom, convicting the innocent, a Type 1 error, is the one the system guards against.
Why is the median better than the mean for income?
Income is highly skewed, so a few very large earners pull the mean upward. The median reflects the typical person and is the better summary.
Practice. Improve. Repeat.