Statistical Inference Questions for SSC CGL

Core Concept: What is Statistical Inference?

Statistical inference is the process of drawing conclusions about a population based on sample data. You observe a small group (sample) and use that to infer truths about the entire group (population). Think of it like tasting one spoonful of soup to decide if the whole pot is salty. There are two main branches: estimation (finding population parameters) and hypothesis testing (deciding if a claim about the population is true or false).

• Population = entire group; Sample = subset we actually measure
• Parameter = unknown truth about population (e.g., true mean); Statistic = calculated from sample (e.g., sample mean)
• Inference = using statistic to estimate or test the parameter
• Always involves uncertainty and probability

Hypothesis Testing: The Framework

Hypothesis testing is a structured method to decide whether to reject or accept a claim about a population. You start with a null hypothesis (H0: the status quo, usually 'no effect' or 'no difference') and an alternative hypothesis (H1: what you suspect is true). You then collect sample data and calculate a test statistic. If the test statistic falls in the rejection region (determined by significance level alpha), you reject H0. Otherwise, you fail to reject it. Remember: you never 'prove' H0 true; you only fail to find evidence against it.

• H0 (Null) = claim of no effect, no difference, or status quo
• H1 (Alternative) = claim you want to prove (one-tailed: > or <; two-tailed: ≠)
• Significance level (alpha) = probability of rejecting H0 when it's actually true (Type I error); typically 0.05 or 0.01
• Test statistic = standardized value calculated from sample; compared to critical value
• p-value = probability of observing data as extreme as yours if H0 is true; reject if p < alpha
• Decision: if test statistic > critical value (or p < alpha), reject H0; otherwise fail to reject

Types of Tests: Z-test vs T-test

Z-test is used when population standard deviation is known OR sample size is large (n > 30). T-test is used when population standard deviation is unknown AND sample size is small (n < 30). The t-distribution has heavier tails than the normal distribution, so critical values are larger, making it harder to reject H0. In SSC CGL, you'll mostly see z-test for large samples and t-test for small samples. Always check the problem statement: if sigma (population SD) is given, use z; if only sample SD is given and n is small, use t.

• Z-test: use when sigma known or n > 30; critical values from standard normal table
• T-test: use when sigma unknown and n < 30; critical values from t-table (depends on degrees of freedom = n-1)
• One-sample test: test mean of one group against a hypothesized value
• Two-sample test: compare means of two independent groups
• Paired t-test: compare two dependent/related samples (before-after, matched pairs)

Estimation: Confidence Intervals

A confidence interval is a range of values that likely contains the true population parameter. For example, a 95% confidence interval for the mean means: if you repeated your sampling process 100 times and calculated a CI each time, about 95 of those intervals would contain the true population mean. The interval is built as: point estimate ± (critical value × standard error). Wider intervals = more confidence but less precision; narrower intervals = more precision but less confidence.

• Confidence level (e.g., 95%, 99%) = probability that the interval contains the true parameter
• Standard error = standard deviation of the sampling distribution; smaller with larger sample size
• Margin of error = critical value × standard error
• Wider CI = higher confidence; narrower CI = lower confidence
• To reduce margin of error: increase sample size or decrease confidence level

Type I & Type II Errors

Type I error (false positive): rejecting H0 when it's actually true. Probability = alpha (significance level). Type II error (false negative): failing to reject H0 when it's actually false. Probability = beta. Power of test = 1 - beta = probability of correctly rejecting a false H0. In SSC CGL, you may see questions asking: 'What is the probability of Type I error?' Answer: alpha. Or 'How to reduce Type II error?' Answer: increase sample size or increase alpha (but this increases Type I error, so there's a trade-off).

• Type I error (alpha) = rejecting true H0; controlled by setting significance level
• Type II error (beta) = failing to reject false H0; reduced by larger sample size
• Power = 1 - beta = ability to detect a true effect
• Trade-off: decreasing alpha increases beta and vice versa
• Increasing sample size reduces both Type I and Type II errors

Sampling Distribution & Central Limit Theorem

The sampling distribution is the probability distribution of a sample statistic (e.g., sample mean) across all possible samples of the same size. The Central Limit Theorem (CLT) states: if you take many random samples from any population, the distribution of sample means approaches a normal distribution, regardless of the shape of the original population. This is why we can use z-test and t-test even if the population isn't normally distributed (as long as n is large enough). Standard error = sigma / sqrt(n), so larger samples have smaller standard error and more precise estimates.

• Sampling distribution = distribution of a statistic across all possible samples
• Central Limit Theorem: sample means are approximately normal if n is large (usually n > 30)
• Standard error = population SD / sqrt(sample size); decreases as n increases
• Sample mean is an unbiased estimator of population mean
• Confidence interval width shrinks as sample size increases

• Confusing H0 and H1: H0 is always the claim of 'no effect' or status quo; H1 is what you suspect. If the problem says 'test if the mean is greater than 50', then H0: mu = 50 and H1: mu > 50, NOT the reverse.
• Using z-test when you should use t-test: If population SD (sigma) is NOT given and sample size is small (n < 30), you MUST use t-test, not z-test. Many aspirants default to z-test without checking.
• Misinterpreting confidence level: A 95% CI does NOT mean there's a 95% chance the true parameter is in this specific interval. It means if you repeated sampling infinitely, 95% of such intervals would contain the true parameter. The true parameter is fixed; the interval is random.
• Rejecting H0 when p-value equals alpha: Standard rule is reject if p < alpha (strictly less than). If p = alpha exactly, you're at the boundary; most texts say fail to reject in this case.
• Confusing Type I and Type II errors: Type I = false positive (rejecting true H0); Type II = false negative (failing to reject false H0). Memory aid: Type I is like a false alarm; Type II is like missing a real threat.

• H0 vs H1: 'H0 = status quo, H1 = hypothesis I want to prove'. H0 is the boring claim; H1 is the exciting one.
• Type I vs Type II: 'Type I = false positive (like a pregnancy test saying yes when you're not pregnant); Type II = false negative (test says no when you are pregnant)'. Alpha controls Type I.
• Z vs T: 'Z = big sample or sigma known; T = tiny sample or sigma unknown'. If n > 30 or sigma given, think Z. If n < 30 and sigma unknown, think T.
• Confidence interval: 'Point estimate ± margin of error'. The ± part is critical value times standard error. Bigger sample = smaller margin = narrower interval = more precise.
• CLT mantra: 'No matter the shape of the population, sample means are normal if n is large'. This justifies using z and t tests even for non-normal populations.

• SSC CGL Tier-II rarely asks more than 1-2 questions on inference, but they're often tricky. Expect: 'Calculate 95% CI for mean given sample data' or 'Test if mean differs from a value at 5% significance level'. Always show your work: state H0/H1, calculate test statistic, compare to critical value, conclude.
• Critical values are usually provided in the question or a table. For z-test at alpha=0.05 (two-tailed), critical value is ±1.96. For alpha=0.01, it's ±2.58. Memorize these for speed.
• Watch for one-tailed vs two-tailed tests. One-tailed (> or <) uses alpha as-is; two-tailed (≠) splits alpha into alpha/2 on each tail. This changes the critical value.
• Degrees of freedom for t-test = n - 1. Always use this when looking up t-table values. A common mistake is using n instead of n-1.
• If a question asks 'What is the probability of Type I error?' the answer is simply alpha (the significance level). If it asks 'What is the probability of Type II error?' you need beta, which is usually not given directly; instead, you're asked to reduce it by increasing sample size.