The t-distribution is a bell-shaped probability distribution used when estimating a population mean from a small sample. It matters because real data often come from samples where the population standard deviation is unknown. Compared with the standard normal distribution, the t-distribution has heavier tails, which allows for more uncertainty.
This makes confidence intervals wider and hypothesis tests more cautious when sample sizes are small.
The shape of a t-distribution depends on degrees of freedom, usually df = n - 1 for a one-sample mean problem. With low degrees of freedom, the curve is flatter in the center and thicker in the tails than the normal curve. As the sample size increases, the sample standard deviation becomes a better estimate of the population standard deviation, and the t-distribution approaches the standard normal distribution.
This is why t-methods are central tools for confidence intervals and tests about means.
Understanding Statistics: The t-Distribution
The main difficulty is that a sample has two kinds of uncertainty. Its average may differ from the true population average just by chance. Its spread may differ from the population spread too.
When students calculate a standard error using the sample standard deviation, they are using one estimate to judge another estimate. That extra layer of uncertainty is the reason the t method is needed.
A small sample can give an unusually low or high estimate of spread. The t distribution accounts for this by making extreme standardized results less surprising than they would be under a normal model.
A t statistic measures how far a sample result is from a claimed population mean in units of estimated standard error. A value near zero says the observed average is close to the claim compared with the normal amount of sample variation. A large positive or negative value says the gap is large relative to that variation.
For a hypothesis test, the location of the t statistic on the relevant t curve helps determine a p value. For a confidence interval, a critical value sets the distance from the sample average to each endpoint. A higher confidence level needs a larger critical value because it aims to capture the population mean more often over many repeated samples.
T methods work best when observations are collected independently and represent the population of interest. Independence is especially important. If one person’s result affects another person’s result, the apparent sample size can exaggerate how much information is available.
The data should come from a population that is roughly normal when the sample is very small. With a larger sample, t procedures usually handle moderate skew fairly well.
They do not handle strong outliers well, because a single extreme value can change both the average and the sample standard deviation. Students should always inspect a graph of the data, such as a dot plot or histogram, before trusting a calculation.
The t distribution appears in many ordinary comparisons. A class might estimate the mean time students spend on homework. A clinic might compare the average blood pressure of two treatment groups.
In a paired study, such as measuring each athlete before and after training, the analysis focuses on the average of the individual differences. This pairing can reduce noise because each person is compared with themselves. It is important to identify the design before choosing a procedure.
One sample, two independent samples, and paired samples use related t methods, but their standard errors reflect different sources of variation. The key habit is to connect the calculation to the data collection process, not merely select a formula from a list.
Key Facts
- For a one-sample mean, degrees of freedom are df = n - 1.
- The t statistic for a sample mean is t = (xbar - μ0) / (s / sqrt(n)).
- Use the t-distribution when σ is unknown and the sample standard deviation s is used.
- A confidence interval for a mean is xbar ± t* s / sqrt(n).
- Smaller df gives heavier tails, so critical values t* are larger.
- As df increases, the t-distribution approaches the standard normal distribution.
Vocabulary
- t-distribution
- A family of bell-shaped probability distributions used to model standardized sample means when the population standard deviation is unknown.
- Degrees of freedom
- The number of independent pieces of information used to estimate variability, often df = n - 1 for a one-sample mean.
- t statistic
- A standardized value that measures how many estimated standard errors a sample mean is from a hypothesized population mean.
- Standard error
- The estimated standard deviation of a sampling distribution, equal to s / sqrt(n) for a sample mean.
- Critical value
- A cutoff value from a probability distribution that marks the boundary of a confidence interval or rejection region.
Common Mistakes to Avoid
- Using the normal distribution when σ is unknown and n is small. This is wrong because replacing σ with s adds extra uncertainty that the t-distribution accounts for.
- Forgetting to use df = n - 1 in a one-sample t problem. This is wrong because the sample mean uses one piece of information, leaving only n - 1 independent deviations.
- Thinking all t-distributions have the same shape. This is wrong because the curve changes with degrees of freedom, especially in the tails.
- Using s instead of s / sqrt(n) in the t statistic. This is wrong because the t statistic compares the mean to its standard error, not to the raw sample spread.
Practice Questions
- 1 A sample of n = 16 has mean xbar = 52 and standard deviation s = 8. Test against μ0 = 50 by computing the t statistic.
- 2 A sample of n = 25 has s = 10. Find the standard error of the mean and the degrees of freedom.
- 3 Explain why a t critical value for df = 5 is larger than a t critical value for df = 50 for the same confidence level.