Diagnosing Claims About Sampling Distributions

Why Sampling-Distribution Claims Need Conditions

A claim about a statistic’s sampling distribution is only as reliable as the conditions behind it. Saying that a sample mean is centered at the population mean, that a larger sample has less variability, or that a distribution is symmetric may be correct in one sampling setup and wrong in another. The statistic, the population, and the sampling method all matter.

A sampling distribution describes the values a statistic would take across repeated samples drawn by the same specified method from the same population. It is not the distribution of individual observations, nor is it the result from just one sample. To diagnose a claim, first identify the sampling method and the statistic, then check its center, standard error, and shape separately.

Definition: A statistic’s sampling distribution is the probability distribution of that statistic over all samples of a fixed size selected according to a specified sampling procedure. Its standard error (\(SE\)) is the standard deviation of this sampling distribution. The standard error describes the statistic’s variability across repeated samples, not the variability of individual observations.

A Diagnostic Framework

For a sample mean \(\bar{x}\), if the observations are independent draws from the same population with mean \(\mu\) and finite variance \(\sigma^2\), its expected value is \(\mu\) and its standard error is \(\sigma/\sqrt{n}\). For a sample proportion \(\hat{p}\), if each observation is an independent Bernoulli trial with success probability \(p\), its expected value is \(p\) and its standard error is \(\sqrt{p(1-p)/n}\). These statements do not automatically apply unchanged to every sample: sampling without replacement from a finite population induces dependence and calls for a finite population correction.

The formulas provide useful checks, but they do not by themselves determine the exact shape. A sampling distribution can be discrete, skewed, or symmetric. For a sample mean, the population’s values and the sample size both affect the shape. For a proportion, the possible values are multiples of \(1/n\), and a success probability near 0 or 1 can make the distribution notably asymmetric. A formula for the standard error is not proof that the sampling distribution is normal.

State the statistic and sampling method.
Specify what is sampled, the sample size, and whether draws are independent or made without replacement.

Check the center.
Identify the population parameter the statistic is intended to estimate and verify the conditions under which its expected value equals that parameter.

Check the standard error.
Use a formula that matches the sampling method. For sampling without replacement, account for the finite population correction.

Assess shape separately.
Use exact probabilities when available, or justify any approximation with conditions appropriate to the situation. Do not infer shape from the center or standard error alone.

Why the Center and Standard Error Formulas Work

A brief calculation explains the usual formulas for an independent sample mean. Let \(X_1,\ldots,X_n\) be independent observations from a population with mean \(\mu\) and variance \(\sigma^2\), and let \(\bar{X}=(X_1+\cdots+X_n)/n\). Linearity of expectation gives the center:

E(\bar{X})=\frac{1}{n}\sum_{i=1}^{n}E(X_i) =\frac{1}{n}(n\mu)=\mu.

Independence makes the variance of the sum equal to the sum of the variances. Therefore:

\operatorname{Var}(\bar{X}) =\frac{1}{n^2}\sum_{i=1}^{n}\operatorname{Var}(X_i) =\frac{n\sigma^2}{n^2} =\frac{\sigma^2}{n}, \qquad SE(\bar{X})=\frac{\sigma}{\sqrt{n}}.

This derivation also exposes an important condition: the variance calculation used independence. When a sample is drawn without replacement from a finite population, its observations are dependent, so the independent-draw standard error needs adjustment.

Worked Example: Checking Center, Standard Error, and Symmetry

Suppose a hypothetical population consists of values 0 and 10, with probability 0.75 of drawing 0 and probability 0.25 of drawing 10. Four observations are drawn independently with replacement. Evaluate the claims that the sample mean is centered at 2.5, has standard error about 2.165, and has a symmetric sampling distribution.

Conditions and center. The draws are independent and come from the same population. The population mean is \(0(0.75)+10(0.25)=2.5\), so the independent-sample result gives \(E(\bar{X})=2.5\). The claim about the center is correct.

Standard error. The population variance is \(0.75(0-2.5)^2+0.25(10-2.5)^2=18.75\), so \(\sigma=\sqrt{18.75}\approx4.330\). Thus:

SE(\bar{X})=\frac{\sigma}{\sqrt{n}} =\frac{\sqrt{18.75}}{\sqrt{4}} \approx 2.165.

The standard error claim is correct. It is in the same units as the population values, here the units of the 0 and 10 measurements.

Shape. The sample mean is \(10K/4\), where \(K\) is the number of 10s in the four draws. In particular, \(\bar{X}=0\) when all four draws are 0, so \(P(\bar{X}=0)=0.75^4=0.31640625\). Also, \(\bar{X}=5\) when exactly two draws are 10, so:

P(\bar{X}=5) =\binom{4}{2}(0.25)^2(0.75)^2 =6(0.0625)(0.5625) =0.2109375.

The values 0 and 5 are equally far from the center 2.5, but their probabilities differ. A distribution symmetric about 2.5 would assign equal probabilities to these equally distant values. Therefore this sampling distribution is not symmetric, and the symmetry claim is false.

Finite Populations and Changing Sample Size

When \(n\) units are selected by simple random sampling without replacement from a finite population of \(N\) values, the observations are dependent: after one unit is selected, it cannot be selected again. If \(\sigma_P^2\) is the variance of the \(N\) population values calculated with divisor \(N\), the sample mean still has expected value equal to the population mean. Its standard error is:

SE(\bar{X})= \frac{\sigma_P}{\sqrt{n}} \sqrt{\frac{N-n}{N-1}}.

The square-root factor is the finite population correction. It is less than 1 when \(n>1\), and it is 0 when the entire population is sampled. Thus, the independent-draw formula alone overstates variability for a sample without replacement from a finite population.

Worked Example: Checking a Claim Under Sampling Without Replacement

A hypothetical finite population has values 2, 4, and 8. Two values are selected by simple random sampling without replacement. A claim says that the sample mean is centered at the population mean and that its standard error equals the independent-draw formula \(\sigma_P/\sqrt{2}\). Check both parts.

Center. The population mean is \((2+4+8)/3=14/3\). The three possible samples are \(\{2,4\}\), \(\{2,8\}\), and \(\{4,8\}\), each with probability \(1/3\). Their means are 3, 5, and 6, so the expected sample mean is \((3+5+6)/3=14/3\). The claim about the center is correct.

Standard error. The population variance, using divisor \(N=3\), is:

\sigma_P^2 =\frac{(2-14/3)^2+(4-14/3)^2+(8-14/3)^2}{3} =\frac{56}{9}, \qquad \sigma_P=\frac{\sqrt{56}}{3}\approx2.494.

The independent-draw formula would give \(2.494/\sqrt{2}\approx1.763\), but this ignores the dependence from sampling without replacement. The finite-population formula gives:

SE(\bar{X}) =\frac{\sqrt{56}/3}{\sqrt{2}} \sqrt{\frac{3-2}{3-1}} =\sqrt{\frac{14}{9}} \approx1.247.

This agrees with the exact distribution: its mean is \(14/3\), and the variance of the equally likely values 3, 5, and 6 is \(\big[(3-14/3)^2+(5-14/3)^2+(6-14/3)^2\big]/3=14/9\). The uncorrected standard-error claim is false.

For independent observations, increasing \(n\) reduces the standard error of a mean or proportion in proportion to \(1/\sqrt{n}\), when the population distribution or success probability stays fixed. In particular, quadrupling \(n\) halves the standard error. Under sampling without replacement, the finite population correction also changes as the fraction sampled changes. In either setting, a smaller standard error means less variability in the statistic across repeated samples; it does not mean that every individual observation is less variable.

Worked Example: Diagnosing a Sample-Size Claim for a Proportion

In a hypothetical process, each independently selected item has probability \(p=0.60\) of meeting a stated criterion. Compare the sampling distributions of \(\hat{p}\) for samples of size 25 and 100. Evaluate the claims that their centers are the same, that the second standard error is half the first, and that the larger sample guarantees a normal shape.

Conditions and center. The trials are independent Bernoulli observations with the same success probability. The sample proportion is the mean of the 0-or-1 success indicators, so \(E(\hat{p})=p=0.60\) for either sample size. The center claim is correct.

Standard errors. For independent Bernoulli observations, the variance of an indicator is \(p(1-p)\), giving \(SE(\hat{p})=\sqrt{p(1-p)/n}\). Substituting the stated values:

SE_{25}=\sqrt{\frac{0.60(0.40)}{25}} =\sqrt{0.0096} \approx0.0980, \qquad SE_{100}=\sqrt{\frac{0.60(0.40)}{100}} =\sqrt{0.0024} \approx0.0490.

The second standard error is half the first, so that claim is correct. Both standard errors are measured in proportion units, or equivalently percentage points if multiplied by 100.

Shape. Under these conditions, the number of successes has a binomial distribution and \(\hat{p}\) is that count divided by \(n\). It takes discrete values, not every value in an interval. The larger sample does not, by itself, prove that the distribution has an exactly normal shape. The claim of a guaranteed normal shape is not justified by the center and standard-error calculations.

Common Mistakes and a Practical Check

A frequent error is to treat four separate claims as if one formula established all of them. Knowing that a statistic is unbiased tells us about its expected center, not its shape. Knowing its standard error tells us how spread out its sampling distribution is, not whether it is symmetric or normal. Likewise, a larger sample can reduce the standard error while leaving the exact distribution discrete.

Another pitfall is to compare probabilities at values that are not equally far from a proposed center when testing symmetry. For a distribution symmetric about \(c\), corresponding values \(c-d\) and \(c+d\) must have matching probabilities (with the appropriate interpretation for a discrete distribution). In the first example, comparing 0 and 10 would not check symmetry about 2.5 because those values are 2.5 and 7.5 units from the center. Comparing 0 and 5 is valid because each is 2.5 units from 2.5, and their unequal probabilities establish asymmetry.

Before accepting a claim, check these points:

Does the stated sampling method support the claimed center?
Are observations independent, or is sampling without replacement relevant?
Does the standard-error formula match the design and use the correct units?
Is shape established from the population and sampling conditions, or merely assumed?
When sample size changes, are the population and sampling method otherwise held fixed?

Key takeaway: Diagnose center, standard error, shape, and sample-size behavior as distinct properties. A correct statement about one does not automatically establish the others; each must be supported by the population model and sampling design.

Check Your Understanding

For each question, identify the relevant sampling conditions before deciding whether the claim is justified.

Independent observations have population mean 12 and population standard deviation 8. What are the expected value and standard error of \(\bar{x}\) for \(n=16\)?
A sample of 20 is drawn without replacement from a population of 40. Why should the independent-draw standard-error formula be checked before using it?
A sample proportion is based on independent Bernoulli trials with \(p=0.10\). Does the formula for its center alone establish that its sampling distribution is symmetric? Explain.
If the sample size for an independent sample proportion is multiplied by 9 while \(p\) stays fixed, by what factor does its standard error change?
To check symmetry about 4, which pair is more relevant: values 1 and 6, or values 1 and 7? Explain.