From a Percentile to a Value
In Finding Percentiles from a Normal Model, you started with a value and found the percentage of the model at or below it. Now reverse that direction: start with a percentage and find the value that marks the cutoff. For instance, a question might ask what measurement separates the highest 10% of values from the other 90%.
The calculator function \(\operatorname{invNorm}\) finds a value from a cumulative area. Its area input is the proportion to the left of the value, not the proportion to its right. This detail matters especially when a question describes the top or highest percentage.
For a normal random variable \(X\sim N(\mu,\sigma)\), a value at the 90th percentile has 0.90 of the model’s area to its left. If the question asks for the cutoff for the top 10%, then 10% of the area is to the right and the remaining 90% is to the left. So the required input is 0.90, not 0.10.
On a TI-84 or similar calculator, enter \(\operatorname{invNorm}(\text{area},\text{mean},\text{standard deviation})\). The area is a decimal between 0 and 1. The mean and standard deviation must use the same units as the measurement you want to find. The result is a value on the original measurement scale, not a z-score.
Choosing the Area Before Using invNorm
First decide which side of the cutoff the problem describes. A percentile is a left-tail area: the 80th percentile, for example, has 0.80 to its left. A “top 10%” cutoff instead describes a right-tail area of 0.10, so its left-tail area is \(1-0.10=0.90\). Sketching a normal curve and marking the requested tail can help prevent reversing these areas.
A normal distribution is continuous, so the area to the left of a cutoff and the area at or below it are the same for this model. The cutoff itself is still a measurement—such as minutes, points, or grams—while the area is a unitless proportion.
Name the measurement \(X\), its units, and the normal model \(N(\mu,\sigma)\).
Decide whether the question gives a left-tail percentile or a right-tail percentage such as “top 10%.”
Use the given percentile as a decimal, or subtract a right-tail proportion from 1.
Enter \(\operatorname{invNorm}(\text{area},\mu,\sigma)\), then state what the resulting value means in context.
The procedure assumes that the stated normal model is appropriate for the variable. It does not establish that a real-world measurement follows a normal distribution; that assumption comes from the problem or from evidence considered elsewhere. Before interpreting the answer, also check that the mean, standard deviation, and requested value use the same measurement units.
Worked Example: The Top 10% of Practice Times
Worked Example: The Top 10% of Practice Times
In a fictional training program, athletes’ times to complete a course are modeled as normal with mean 68 seconds and standard deviation 4 seconds. Find the cutoff for the slowest 10% of times.
State. Let \(X\) be an athlete’s course time, in seconds. The model is \(X\sim N(68,4)\). We want the cutoff with 10% of modeled times above it.
Plan. “Slowest 10%” means the right-tail area is 0.10. Since invNorm needs the area to the left, use \(1-0.10=0.90\). The cutoff will be in seconds.
Do. Enter the left-tail area, mean, and standard deviation in that order:
The calculator’s inverse result corresponds to a standard-normal cutoff of about \(z=1.28155\), or \(1.282\) rounded to three decimal places. As a check on the area, use the unrounded cutoff in the original model: \(\operatorname{normalcdf}(73.1262,1E99,68,4)\approx0.1000\), rounded to four decimal places. This confirms that about 10% of modeled times are above the cutoff.
Conclude. The cutoff for the slowest 10% of course times is about 73.13 seconds. In this model, about 10% of athletes have times greater than this value, and about 90% have times at or below it.
Worked Example: A Lower Cutoff for Delivery Waits
Worked Example: A Lower Cutoff for Delivery Waits
For a fictional delivery service, customer wait times are modeled as normal with mean 32 minutes and standard deviation 6 minutes. Find the 15th percentile of wait times and interpret it.
State. Let \(X\) be a customer’s wait time, in minutes. The model is \(X\sim N(32,6)\). The 15th percentile is the value with 0.15 of the model’s area to its left.
Plan. This is a left-tail percentile, so use 0.15 directly as the area input. Because 15% is below half of the distribution, the cutoff should be below the mean of 32 minutes.
Do. The inverse-normal calculation is:
The corresponding z-score is about \(-1.0364\). A check with the original model gives \(\operatorname{normalcdf}(-1E99,25.7814,32,6)\approx0.1500\), rounded to four decimal places. The area to the left agrees with the 0.15 input.
Conclude. The 15th-percentile wait time is about 25.78 minutes. About 15% of modeled customer wait times are at or below this value, and about 85% are above it. The cutoff is below the mean, as expected for a percentile below 50.
Worked Example: A Score at the 95th Percentile
Worked Example: A Score at the 95th Percentile
A fictional placement assessment has scores modeled as normal with mean 500 points and standard deviation 100 points. Find the score at the 95th percentile.
State. Let \(X\) be an assessment score, in points. The model is \(X\sim N(500,100)\), and the requested value has 0.95 of the modeled scores to its left.
Plan. The 95th percentile is a left-tail percentile, so enter 0.95 directly. Since 0.95 is greater than 0.50, the cutoff should be above the mean.
Do. Use the normal model’s mean and standard deviation as the second and third inputs:
The standardized cutoff is approximately \(z=1.6449\). To check the percentile using the original units, calculate \(\operatorname{normalcdf}(-1E99,664.4854,500,100)\approx0.9500\), rounded to four decimal places. The result is greater than 500 points, consistent with a percentile above 50.
Conclude. The score at the 95th percentile is about 664.49 points. About 95% of modeled assessment scores are at or below this value, while about 5% are above it.
Interpreting and Checking an Inverse-Normal Result
The invNorm result is a cutoff in the original units. It is not the area and is not a z-score. In the practice-time example, 73.1262 is measured in seconds; 0.90 was the left-tail proportion entered into the calculator; and approximately 1.282 describes the cutoff’s distance above the mean in standard-deviation units.
A useful check is to return to the original question and verify the side and size of the tail. For the top 10%, the cutoff must be above the mean, and only 10% of the model should lie above it. For a low percentile such as the 15th, the cutoff should be below the mean, with 15% to its left. You can check a calculated cutoff with normalcdf using the appropriate tail and the model’s original parameters.
Calculator output may include more digits than are useful in a final answer. Keep the unrounded value for checking, then round the cutoff to a sensible precision for the context. If a measurement is recorded to the nearest whole point or minute, a final answer may be rounded to that unit; explain the model-based cutoff clearly rather than implying more precision than the setting supports.
Common Mistakes and AP Exam Tips
- Entering the top percentage directly. For the top 10%, entering 0.10 finds a cutoff with 10% to its left, not 10% to its right. Convert the requested right-tail proportion to a left-tail area: \(1-0.10=0.90\).
- Entering a percent instead of a proportion. The area input is a decimal such as 0.90, not 90. Convert the percentage before using the calculator.
- Reversing the calculator inputs. The order is area, mean, standard deviation. A correct area with the mean and standard deviation switched will not answer the problem.
- Reporting an area as the cutoff. A result such as 0.90 is a proportion, not a measurement. invNorm returns the cutoff in the original units.
- Giving a z-score instead of the requested value. A z-score can help check a result, but the question asks for the original-scale cutoff. Use the actual mean and standard deviation with invNorm.
- Leaving out context in the conclusion. A complete interpretation names the measurement, gives the cutoff with units, and says what proportion of modeled values lies above or below it.
For a full-credit response, define \(X\) and its normal model, identify whether the stated percentage is a left-tail percentile or a right-tail proportion, show the area conversion when needed, and write the invNorm calculation with its inputs. Conclude in context and check that the cutoff’s position relative to the mean makes sense.
Key Takeaway
To find a value from a percentile in a normal model, enter the left-tail area, mean, and standard deviation into invNorm. A stated percentile is already a left-tail percentage; a “top” percentage must first be converted to its complementary left-tail area. The calculator returns the cutoff in the variable’s original units.
Check Your Understanding
For each question, identify the area input to invNorm and describe what the result represents.
- A fictional greenhouse models plant heights as \(N(24,3)\), in centimeters. Write the invNorm command for the cutoff marking the tallest 5% of plants.
- For \(X\sim N(80,12)\), write the invNorm command for the 30th percentile. Should the cutoff be above or below the mean?
- A normal model for package weights has mean 2.4 kilograms and standard deviation 0.3 kilograms. Explain why the 90th percentile uses 0.90 as the area input rather than 0.10.
- A student enters \(\operatorname{invNorm}(0.08,40,5)\) to find the cutoff for the top 8% of modeled values. Identify the error and give the correct area input.
- Explain why an invNorm result is reported in the measurement’s units rather than as a percentage or a z-score.