Statistics interview questions

When the data is skewed or has outliers, because the mean gets pulled toward the extreme values. For example, if most customers spend around 500 but a few spend 50,000, the mean spend would overstate what a typical customer does. I usually report both and explain the gap.

I would say it shows how surprising our result would be if there were really no difference. With p = 0.02, a difference this large would rarely happen by chance alone, so we treat it as a real effect. I also add how large the difference is, because significant does not always mean important.

A Type I error is a false positive: concluding there is an effect when there is none, like launching a feature that does not actually help. A Type II error is a false negative: missing a real effect, often because the sample was too small. Lowering one risk usually raises the other unless you collect more data.

If we repeated the sampling many times and built an interval each time, about 95% of those intervals would contain the true value. In practice I present it as a range of plausible values, for example average delivery time between 48 and 52 minutes. A narrow interval means a precise estimate.

First I look for a third factor that could drive both, such as season, region or marketing spend, and check whether the relationship holds after splitting by it. The strongest evidence comes from a controlled test, such as an A/B test where only one factor changes. Without that, I describe it as an association, not a cause.

Larger samples reduce the standard error, which shrinks with the square root of n, so estimates become more precise and real effects are easier to detect. But size does not fix bias: a large survey of only app users still misrepresents customers who do not use the app. I check both how many and who was sampled.