Population Shape, Sample Size, and Mean Distributions

Why Look at Both Population Shape and Sample Size?

A population plot describes individual values. A sampling distribution of the sample mean describes something different: the values of $\bar{x}$ that would result from repeatedly taking samples of the same size under the same sampling model. A population may have a long tail, but each sample mean combines several observations. Looking at the two distributions together helps us understand what averaging changes—and what it does not necessarily remove.

Repeated-sample simulations can make this comparison visible. They also guard against a tempting overstatement: a larger sample often makes a sample-mean distribution look less skewed, but sample size alone does not guarantee a particular shape for every population or sampling procedure.

Definition: The population distribution describes the values of an individual measurement in the population. The sampling distribution of $\bar{x}$ describes the values of the sample mean across all possible repeated samples of a specified size, taken under a specified sampling model. A simulation produces only a finite set of those possible sample means; its plot is an approximation to the sampling distribution.

What Population Skewness Can Carry Into Sample Means

A distribution is right-skewed when its right tail extends farther than its left tail; a few unusually large values may sit well above the bulk of the data. A distribution is left-skewed when its left tail extends farther. A roughly symmetric distribution has similar shapes on either side of its center. These descriptions concern the distribution’s shape, not whether any one sample happens to include an extreme value.

When the population is roughly symmetric, sample means often have a roughly symmetric distribution as well. When the population is strongly skewed, the distribution of sample means for small samples can retain visible skewness. As sample size grows, individual extreme observations have less influence on an average, and repeated-sample means commonly become more concentrated and less visibly skewed. These are useful tendencies, not guarantees or a rule that says a particular sample size makes every distribution normal.

Population plot	Smaller sample sizes	Larger sample sizes
Roughly symmetric	Sample means are often roughly symmetric, though they may be discrete or irregular.	Sample means are often more concentrated; their general balance may remain visible.
Right- or left-skewed	Sample means may retain the population’s tail direction, especially when samples are small.	Skewness often becomes less pronounced, but a tail or irregular shape may remain.
Strongly skewed or with rare extremes	Results can vary substantially depending on whether a sample includes an extreme value.	More averaging may reduce the effect of any one observation, but there is no universal sample-size cutoff for a desired shape.

The population mean $\mu$ is the center to compare against. Under the sampling model used for the earlier results on sample means, the sampling distribution is centered at $\mu$, and its standard error is $SE=\sigma/\sqrt{n}$. That standard error describes the distribution’s spread; it does not describe its skewness. Two sampling distributions can have the same standard error but different shapes.

A Short Justification: Why One Extreme Value Has Less Leverage

The sample mean is the sum of the $n$ observations divided by $n$. Relative to a population mean $\mu$, this gives

\bar{x}-\mu=\frac{1}{n}\sum_{i=1}^{n}(x_i-\mu).

If one observation is $d$ units above $\mu$, while the other $n-1$ observations are exactly equal to $\mu$, then the sample’s mean deviation is $d/n$. For example, a single observation 20 units above $\mu$ contributes 20 units to the deviation when $n=1$, but 5 units when $n=4$. This establishes how averaging dilutes that one observation’s contribution in this situation. It does not prove that all sample-mean distributions must have a particular shape: the other observations also vary, and repeated samples can include more than one extreme value.

Compare the Shapes Using Repeated Samples

Worked Example: A Symmetric Population and Pairs of Observations

Consider a hypothetical population whose possible values are 2, 4, 6, and 8, each equally likely. The population plot is symmetric around 5. Its mean is

\mu=\frac{2+4+6+8}{4}=5.

For a small, exact illustration, suppose each possible ordered pair in a sample of size $n=2$ is equally likely. There are $4\times4=16$ pairs. Their sample means have the following frequencies.

Sample mean $\bar{x}$	Frequency out of 16
2	1
3	2
4	3
5	4
6	3
7	2
8	1

For instance, a mean of 3 comes from the pairs $(2,4)$ and $(4,2)$, while a mean of 5 comes from $(2,8)$, $(4,6)$, $(6,4)$, and $(8,2)$. The frequencies mirror one another around 5. Their weighted mean is

\frac{2(1)+3(2)+4(3)+5(4)+6(3)+7(2)+8(1)}{16} =\frac{80}{16}=5.

In this example, the population is symmetric and the sample-mean distribution is symmetric around the same center. This small, discrete case illustrates a pattern; it does not establish that every sample-mean distribution will be smooth or bell-shaped.

Worked Example: A Right-Skewed Population With Four-Observation Means

Suppose a hypothetical population of service times, in minutes, consists of the eight values 1, 1, 1, 2, 2, 3, 4, and 8, each equally represented in the population. Most values are low, with one much larger value producing a right tail. The population mean is

\mu=\frac{1+1+1+2+2+3+4+8}{8} =\frac{22}{8}=2.75\text{ minutes}.

A simulation of repeated samples of size $n=4$ produced these 16 sample means, in minutes:

1.50, 1.75, 2.00, 2.25, 2.50, 2.50, 2.50, 2.75, 2.75, 3.00, 3.00, 3.25, 3.25, 3.50, 4.00, 5.50.

Each value is calculated by adding the four observations in one simulated sample and dividing by 4. For example, a sample with observations 8, 8, 4, and 2 has mean $(8+8+4+2)/4=5.50$ minutes. Grouping the 16 simulated means shows that most fall between 2.50 and 3.25 minutes, while one is far out at 5.50. The sample-mean plot is more concentrated than the population values, but this small simulation still has a long right tail.

The average of the 16 simulated means is $46/16=2.875$ minutes, not exactly 2.75. That difference is plausible: a finite simulation is only a sample of possible sample means, so its average need not equal the sampling distribution’s center. The simulation illustrates both a tendency toward concentration and the possibility that skewness remains visible.

Worked Example: A Rare Extreme Value Can Matter at a Larger Sample Size

Consider a hypothetical population in which an observation is 0 minutes with probability 0.9 and 10 minutes with probability 0.1. The population mean is

\mu=0(0.9)+10(0.1)=1\text{ minute}.

A sample of size $n=1$ has a strongly right-skewed distribution: 90% of individual values are 0, while 10% are 10. For samples of size $n=10$ under this specified model, let $K$ be the number of observations equal to 10. The sample mean is $K$ minutes, because the sample total is $10K$ minutes and $10K/10=K$.

For example, the probability of getting no 10-minute observations is $0.9^{10}\approx0.3487$. The probability of exactly one is

\binom{10}{1}(0.1)(0.9)^9 =10(0.1)(0.9)^9 \approx0.3874.

The probability of exactly two is

\binom{10}{2}(0.1)^2(0.9)^8 =45(0.01)(0.9)^8 \approx0.1937.

Thus about $0.3487+0.3874+0.1937=0.9298$ of the sample means are 0, 1, or 2 minutes. The remaining probability, about $1-0.9298=0.0702$, lies above 2 minutes, forming a right tail. The distribution has several possible means rather than only 0 and 10, yet it is not symmetric. A larger sample has changed the possible values and concentrated much of the probability near the mean; it has not erased the right tail.

Reading Simulation Plots Carefully

When comparing repeated-sample plots, first check that they use the same population and sampling procedure. Then look at their centers, spreads, and shapes separately. The center concerns where the sample means cluster; the spread concerns how much they vary; the shape concerns such features as symmetry, tails, gaps, and isolated values. A plot that is narrower is not automatically more symmetric.

Also separate the theoretical sampling distribution from its simulated approximation. A simulation with few repetitions can look lopsided just by chance, or fail to show a rare tail event. With more repetitions, the display may better reveal the underlying pattern, but it does not change the population or the sample size used to calculate each mean. When discussing results, identify whether a statement refers to the population, the sampling distribution, or the finite simulation.

Key takeaway: Population skewness is most likely to remain conspicuous in sample-mean distributions when samples are small or extreme values are influential. Increasing sample size generally concentrates means and often softens skewness, but neither symmetry nor a particular bell-shaped form is guaranteed by a universal cutoff.

Common Mistakes

Assuming a large sample guarantees a normal-shaped distribution. A larger sample often reduces skewness, but this is a tendency, not an unconditional guarantee about every population or sampling design.
Confusing spread with shape. The standard error describes the spread of sample means. It does not say whether the distribution is symmetric or skewed.
Calling one simulated plot the exact sampling distribution. A finite simulation displays only the sample means generated in that run. Random variation can make its apparent shape imperfect.
Assuming the sample means in a simulation must average exactly to $\mu$. The sampling distribution may be centered at $\mu$, while the average of a finite set of simulated means differs from $\mu$.
Rounding too early. For example, if $\sigma^2=4.1875$ and $n=10$, retain the exact variance through the standard-error calculation: $SE=\frac{\sqrt{4.1875}}{\sqrt{10}} =\sqrt{\frac{4.1875}{10}} =\sqrt{0.41875} \approx0.6471.$ The result is about 0.6471 units, not 0.6470; rounding $\sigma$ before dividing can change the final digit.

Check Your Understanding

Use the distinctions between population shape, sample size, and sampling-distribution shape in your answers.

A population plot is strongly left-skewed. What shape might you expect for means from small samples, and what may change as sample size grows?
Why does a narrower sample-mean distribution not by itself show that the distribution has become more symmetric?
A simulation produces 20 sample means whose average differs slightly from $\mu$. Does that alone show the sample mean is biased? Explain.
In the rare-extreme example, what value of $K$ gives a sample mean of 3 minutes when $n=10$?
Why should a simulation with only a few repetitions be interpreted cautiously when the population has rare extreme values?