Why Check the Sample’s Shape?
A one-sample t interval uses the sample mean \(\bar{x}\) to estimate a population mean \(\mu\), with uncertainty measured using the sample standard deviation \(s\). Both summaries can be affected by the shape of the data. A long tail or an unusual observation may pull the mean and change the estimated spread, so a numerical interval should not be separated from a look at the observations.
A plot cannot prove that a population is normal, and no single visual rule decides every case. The practical question is narrower: does the sample show strong skewness or an influential outlier that makes a t interval’s approximation questionable? This tutorial focuses on using plots to answer that question. Random sampling, independence, and the 10% condition are separate requirements; a shape plot does not check them.
For an actual interval, the sampling plan must also support treating the observations as independent and representative of the target population. Those conditions cannot be established from a dotplot, histogram, or normal probability plot. Here, assume the sampling conditions have been checked separately, then use the graphs to inspect shape and unusual values.
What to Look for in a Plot
A dotplot displays each observation as a dot along a numerical scale. It is especially useful when the sample is small enough that individual values remain visible. Look for a roughly balanced pattern, gaps, a long tail, or a point well separated from the rest.
A histogram groups observations into bins and shows how many fall in each bin. It can make an overall pattern—such as a long right tail—easy to see. But the appearance depends partly on bin widths and boundaries. A histogram can hide individual values, so a dotplot or a check of the raw data can be useful when a bar suggests an unusual observation.
A normal probability plot (also called a normal Q–Q plot) compares ordered sample observations with the values expected at corresponding positions under a normal distribution. If the points follow an approximately straight pattern, the sample is broadly compatible with normal shape. Systematic curvature suggests skewness; a point that departs markedly from the general line may signal an unusual observation. The plot is evidence, not a pass/fail test.
Interpret the graphs in context. A value can be unusual because it reflects a recording error, because it comes from a different population or process, or because the population genuinely has a rare but valid value. A surprising value should be checked against the source and the study design. It should not be deleted merely because it makes an interval less convenient.
Worked Example: A Dotplot Without a Strong Shape Warning
Suppose a hypothetical sample records minutes spent waiting at a service desk for 12 randomly selected visits. The observed times are \(8, 9, 9, 10, 10, 10, 11, 11, 12, 12, 13, 13\). Assume the random-sampling and independence conditions have been checked separately. What does the sample’s shape suggest?
Step 1: Inspect the individual values. A dotplot would place one dot at 8, two at 9, three at 10, two at 11, two at 12, and two at 13. The values occupy a continuous range from 8 to 13 minutes; there is no isolated point far from the cluster and no conspicuously long tail.
Step 2: Check the center pattern. The observations are concentrated near 10–12 minutes, with observations on both sides. They are not perfectly symmetric, but the dotplot gives no clear evidence of strong skewness or an influential outlier.
Step 3: State what the plot does and does not support. On shape grounds, this sample provides no obvious warning against a one-sample t interval. That is not proof that the population is normal, nor does the plot establish the sampling conditions. It is a reasoned assessment of the observed values, in context, rather than a guarantee about unseen observations.
Skewness and Outliers Can Affect the Mean
Skewness describes asymmetry in a distribution. In a right-skewed distribution, most observations may be moderate while a smaller number extend far to the right; left skewness extends farther to the left. Because the mean uses every value, a long tail can pull it toward the tail. The sample standard deviation also responds to extreme distances from the mean.
An outlier is an observation that appears unusually far from the rest of the data. There is no universal numerical cutoff that makes an observation an outlier for every variable and context. In this setting, the practical concern is whether an observation is both unusual and influential for the proposed interval. “Influential” means that including the value materially changes the sample summary or resulting interval; it is a judgment that depends on the data and context.
Here \(M\) is one additional observation and \(\bar{x}_{\text{new}}\) is the mean after adding it. To justify the result, the sum of the original \(n\) values is \(n\bar{x}\), by the definition of the sample mean. Adding \(M\) gives a total of \(n\bar{x}+M\) across \(n+1\) observations, so division by \(n+1\) gives the first equation. Subtracting \(\bar{x}\) yields
Thus, an observation far from the original mean can move the mean. This identity describes the change in the mean; it does not by itself determine whether an observation is erroneous or should be removed. The context and the reason for the value matter.
Worked Example: Right Skew and a Long Wait
Suppose a hypothetical sample contains waiting times, in minutes, of \(2, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 7, 8, 12, 25\). Assume the sampling conditions have been assessed separately. Use the values to evaluate the shape concern.
Step 1: Summarize the pattern. Most observations are between 2 and 8 minutes, while 12 and especially 25 extend the upper tail. A dotplot would show a cluster at the lower values and a separated high value; a histogram would show a long right tail. The median is 5 minutes: the eighth of the 15 ordered observations is 5.
Step 2: Calculate the mean. The total is \(2+3+3+4+4+4+5+5+5+6+6+7+8+12+25=99\). Therefore,
Step 3: Evaluate the high observation’s influence. If 25 minutes is temporarily left out for this diagnostic comparison, the remaining total is \(99-25=74\) minutes across 14 observations, giving \(74/14=5.286\) minutes. The comparison shows that the 25-minute wait pulls the mean upward. It does not justify deleting that observation: first check whether it is a valid wait and belongs to the target population.
Conclusion. This sample shows clear right skew and a potentially influential high value. Those are meaningful cautions about using a t interval. Whether the value is a recording error, a valid rare wait, or evidence about the population must be resolved from context, not from the desired result.
Using a Normal Probability Plot
A normal probability plot is useful because it checks the pattern across the ordered observations rather than relying only on the apparent shape of histogram bars. Rough alignment with a straight line supports compatibility with normal shape. Curvature across many points suggests a systematic departure; a single point far from an otherwise straight pattern calls for investigation.
The plot is not a formal test that proves or disproves normality. Small samples can make patterns difficult to distinguish, and large samples can make minor departures visible. Therefore, read the plot together with a dotplot or histogram, the raw observations, and knowledge of how the measurements were obtained.
Worked Example: A Point Away From the Normal Pattern
A hypothetical sample of 10 measurements, in millimetres, is \(14, 15, 16, 16, 17, 18, 18, 19, 20, 31\). Suppose a normal probability plot shows the first nine ordered values following an approximately straight pattern while the point for 31 mm sits well above the line. What should be concluded about a t interval?
Step 1: Check the arithmetic and center. The first nine values sum to \(14+15+16+16+17+18+18+19+20=153\), so their mean is \(153/9=17.0\) mm. Including 31 mm gives a total of \(184\) mm and a mean of
Step 2: Quantify the change in the mean. By the mean-update formula, adding 31 mm to the first nine observations changes the mean by \((31-17.0)/(9+1)=14/10=1.4\) mm. The result agrees with the direct calculation: the mean rises from 17.0 to 18.4 mm.
Step 3: Interpret the plot and the comparison. The point’s departure from the general line and its effect on the mean make 31 mm a plausible influential observation. Check the measurement record and the population definition. If it is a valid observation from the intended population, removing it without justification would change the data being analyzed. If it is a recording error, correction or exclusion should follow a defensible documented rule.
Conclusion. As recorded, the sample gives a clear shape warning, so a t interval should not be used mechanically. The plot does not decide whether the value is valid, and the mean comparison alone does not settle whether an interval is appropriate. The measurement context and sampling design remain essential.
A Practical Review Before Calculating
A useful review is to look first at individual values and then at the overall pattern. A dotplot preserves detail; a histogram highlights broad shape; a normal probability plot helps assess how closely the ordered observations follow a normal pattern. When these displays tell different stories, inspect the raw data and consider whether histogram bin choices explain the difference.
Check that the data are quantitative measurements of the quantity whose population mean is the target.
Random sampling, independence, and the 10% condition concern how observations were obtained; a shape plot cannot verify them.
Use a dotplot or histogram to look for strong skewness, gaps, clusters, and unusually distant observations. A normal probability plot can add evidence about departures from a straight-line pattern.
Check records and context before deciding whether an observation is erroneous or belongs in the analysis. Do not remove valid data just because a plot flags them.
State whether the observed shape provides a concern for the t interval, and explain why. A visual check supports judgment; it does not certify the population distribution.
Common Mistakes and Why the Check Matters
A frequent mistake is to treat a plot as a pass/fail test: “The histogram looks roughly bell-shaped, so all assumptions are satisfied.” A histogram only describes the observed sample, and its appearance depends on binning. The correction is to use more than one relevant display where helpful, inspect individual observations, and assess sampling conditions separately.
Another mistake is to remove the most distant value automatically. That can conceal genuine variation and alter the population being represented. Instead, verify the data and define a defensible treatment based on the measurement and study context. If the value is valid, report the shape concern rather than silently discarding it.
This check matters because a t interval is meant to quantify uncertainty about a population mean, not merely produce endpoints from a formula. A conspicuous tail or influential point can make the calculation difficult to interpret as a reliable estimate. Conversely, a plot with no strong warning is useful evidence, but it is not proof that every assumption holds.
Check Your Understanding
Use the questions to review how graphical checks inform a one-sample t interval.
- What features in a dotplot might suggest strong skewness or an unusual observation?
- Why can changing histogram bin widths alter the apparent shape of a sample?
- A normal probability plot is nearly straight except for one point far from the pattern. What should be checked before deciding how to handle that observation?
- In your own words, explain why a shape plot cannot establish random sampling or independence.
- Suppose adding one large observation noticeably increases the sample mean. Does that alone justify removing the observation? Explain.