Read the Interval Using the Difference Definition
In “Constructing a Paired t Confidence Interval,” you learned how to calculate an interval for the population mean difference, \(\mu_d\). This tutorial focuses on what the finished interval says. To interpret it correctly, begin with the definition of each difference: the subtraction order determines what a positive or negative value means.
For example, if \(d_i=\text{after}_i-\text{before}_i\), then positive differences represent higher after measurements, and negative differences represent lower after measurements. A paired t interval estimates the population mean of those after-minus-before differences, not the mean of either measurement by itself and not the change for every individual.
Include the confidence level, both endpoints, the units, the subtraction order, and the population. If the full interval is above zero, then every value in the interval describes a positive population mean difference under the stated order. If it is below zero, every value describes a negative population mean difference. If zero is between the endpoints, the interval includes negative, zero, and positive values for the population mean difference.
These sign descriptions are about the mean difference. An interval entirely above zero does not mean every person or object had a positive difference. Individuals can vary, even when the population mean difference is positive. Similarly, a negative interval does not say that every individual’s measurement decreased.
What “95% Confident” Means
The confidence level describes the long-run performance of the method. If we repeatedly took random samples from the same population and constructed intervals in the same way, about 95% of the resulting 95% confidence intervals would capture the true population mean difference. The population parameter \(\mu_d\) is fixed; it is the intervals that vary from sample to sample.
For one particular interval, a careful interpretation says that we are 95% confident the interval captures \(\mu_d\). Do not say there is a 95% probability that this fixed parameter is inside this particular interval, or that 95% of individuals’ differences fall inside it. A confidence interval estimates a population mean; it is not a range describing individual differences.
Interpret Direction and Size in Context
The signs of the endpoints give a quick direction check, but the context gives those signs meaning. For \(d_i=\text{after}_i-\text{before}_i\), a positive endpoint is a possible positive value of the population mean difference (a higher mean after measurement), and a negative endpoint is a possible negative value (a lower mean after measurement). If the measurement is time, for instance, a positive difference means more time after, not necessarily an improvement. Whether an increase is desirable depends on the situation.
The numerical size matters too. An interval from 0.2 to 0.8 seconds describes a different range of average changes than an interval from 20 to 80 seconds. State the units and name the measured quantity so that a reader can judge what the endpoints represent.
If the subtraction order is reversed, the signs and order of the endpoints change. Suppose the original interval for after minus before is \((L,U)\). The interval for before minus after is \((-U,-L)\): negate each endpoint and put the smaller value first. This keeps the new interval in increasing order.
For example, an interval of \((-4.3,-2.5)\) minutes for after minus before becomes \((2.5,4.3)\) minutes for before minus after. Both intervals describe the same range of mean changes, expressed in opposite directions.
Worked Examples
Worked Example: Daily Reading Time After a Reminder Program
A fictional study randomly selects 30 students from a large district and records each student’s average daily reading time before and after a reminder program. The differences are defined as \(d_i=\text{after}_i-\text{before}_i\), in minutes. A paired t procedure produces the 95% confidence interval \((1.2,\ 4.6)\) minutes. Interpret the interval.
State. Let \(\mu_d\) be the true mean after-minus-before difference in average daily reading time, in minutes, for students in the district. The interval estimates this population mean difference.
Plan. The same students were measured twice, so the observations are paired. We interpret the endpoints according to after minus before, not as two separate estimates for before and after. The study reports a random sample; suppose the sampled students can reasonably be treated as independent and the district contains more than 300 students, so \(30<0.10(300)=30\) would not be sufficient if the population were exactly 300. Here the district is stated to be large enough that the 10% condition is met. The study also reports that the differences have no strong skewness or pronounced outliers. These facts support the paired t procedure.
Do. Both endpoints are positive. The interval goes from 1.2 minutes to 4.6 minutes, so its values describe a positive mean after-minus-before difference. Keep the units: these are minutes of average daily reading time.
Conclude. We are 95% confident that the true mean increase in average daily reading time after the reminder program, compared with before it, is between 1.2 and 4.6 minutes for students in the district. This statement concerns the population mean difference; it does not claim that every student increased their reading time by that amount.
Worked Example: Noise Levels After Sound Panels
A fictional facilities team randomly selects 10 classrooms from 180 similar classrooms and measures the average noise level in each room before and after installing sound panels. Define \(d_i=\text{after}_i-\text{before}_i\), in decibels. The 10 differences have \(\bar d=-3.4\) decibels and \(s_d=1.2\) decibels. A graph shows a roughly symmetric distribution without pronounced outliers. Construct and interpret a 95% paired t interval.
State. Let \(\mu_d\) be the true mean after-minus-before change in average classroom noise level, in decibels, for the population of similar classrooms. We seek a 95% confidence interval for \(\mu_d\).
Plan. Each classroom is measured before and after, giving paired data. The classrooms were randomly selected, and the distinct classrooms can reasonably be treated as independent. The 10% condition is met because \(10<0.10(180)=18\). There are only 10 differences, so we check their shape; the roughly symmetric pattern without pronounced outliers supports using a t procedure. The conditions support a paired t interval.
Do. With \(n=10\) pairs, \(df=10-1=9\). For a 95% interval, the positive critical value is \(t^*=\text{invT}(0.975,9)\approx2.262\). The standard error and margin of error are
The endpoints are the sample mean difference minus and plus the margin of error:
which rounds to \((-4.26,\ -2.54)\) decibels.
Conclude. We are 95% confident that the true mean after-minus-before change in average noise level is between \(-4.26\) and \(-2.54\) decibels for similar classrooms. Because the entire interval is negative, it describes a lower average noise level after installation: the mean decrease is between 2.54 and 4.26 decibels. It does not mean that every classroom became quieter by that amount.
Worked Example: Water Use After a Garden Schedule Change
A fictional neighborhood project records water use for the same 20 gardens before and after a new watering schedule. Differences are defined as \(d_i=\text{after}_i-\text{before}_i\), in liters per day. A 90% paired t interval for the population mean difference is \((-0.8,\ 1.6)\) liters per day. Interpret it in context.
State. Let \(\mu_d\) be the true mean after-minus-before difference in daily water use, in liters per day, for gardens in the target population. The reported interval estimates \(\mu_d\).
Plan. Each garden has a before and an after measurement, so the data are paired and the interval concerns the list of within-garden differences. Interpret the confidence level as a long-run property of the interval method. Since zero lies between the endpoints, the interval includes negative as well as positive possible values for the population mean difference.
Do. The lower endpoint, \(-0.8\) liters per day, represents a possible mean decrease of 0.8 liters per day under the after-minus-before order. The upper endpoint, \(1.6\) liters per day, represents a possible mean increase of 1.6 liters per day. The interval also includes zero, a mean difference of no change.
Conclude. We are 90% confident that the true mean after-minus-before difference in daily water use for gardens in the target population is between a decrease of 0.8 liters per day and an increase of 1.6 liters per day. The interval allows mean differences in either direction, including zero. This interpretation alone does not establish that the schedule changed water use or that it had no effect.
Common Mistakes and AP Exam Tips
An interpretation can be numerically correct but still lose credit if it leaves out the population, units, difference order, or confidence level. Use a sign audit before writing: confirm the lower and upper endpoints, check their positions relative to zero, and translate them using the stated definition of \(d_i\).
- Reversing the meaning of a negative difference. With after minus before, a negative value means the after measurement is lower. State that direction in the specific context rather than merely saying “the result is negative.”
- Claiming every individual changed by an amount in the interval. The interval estimates a population mean difference, not the range of individual differences or a prediction for each person.
- Leaving out the subtraction order. “The mean change is between ...” can be ambiguous. Say “mean after-minus-before change” or give an equally explicit definition.
- Calling the confidence level the probability that this fixed parameter is in this interval. Explain that the method captures the fixed parameter in the stated percentage of repeated samples in the long run.
- Reversing endpoints incorrectly. When changing from after minus before to before minus after, negate and swap the endpoints. An interval \((L,U)\) becomes \((-U,-L)\), not \((-L,-U)\).
- Overstating an interval that includes zero. Say that the interval includes negative, zero, and positive mean differences. Do not turn that description into a claim that the treatment definitely had no effect.
For full-credit communication, write one sentence that identifies the confidence level, interval endpoints, units, population, and the explicitly ordered mean difference. Then say what the signs mean in context and keep the claim about the population mean distinct from claims about individuals.
Check Your Understanding
Use the interval and difference definition in each question to describe what the interval means.
- Differences are defined as after minus before, and a confidence interval is \((2.1,\ 5.4)\) centimeters. What direction of population mean difference does the interval describe?
- For after-minus-before differences, an interval is \((-3.0,\ -0.7)\) minutes. Explain the direction in context without claiming that every individual’s time changed by that amount.
- A 95% interval is \((-1.2,\ 2.5)\) kilograms. What directions of population mean difference does it include?
- In one sentence, explain what “95% confident” means for a paired t interval.
- An interval for after minus before is \((-6,\ -2)\) units. What is the corresponding interval for before minus after?