Home >  Exploring Excel  > 

An Excel Tutorial:

An Introduction to Excel's
Normal Distribution Functions

Excel provides several spreadsheet functions for working with
normal distributions or "bell-shaped curves." Here's a brief
introduction for people who are statistically challenged.

by Charley Kyd
July, 2006

(Update: To learn how to create figures like this in Excel, see How to Create Normal Curves In Excel, With Shaded Areas.)

Recently, a reader asked me how to generate a random number from a Normal distribution. This question set me to thinking about doing statistics with Excel.

Many of us were introduced to statistics in school and then forgot what little we learned...often within seconds of graduation. Also, when we took statistics, many of us weren't taught how to work with statistics using Excel. This is unfortunate, because in business it's often useful to have some grasp of that topic.

For all these reasons, I thought it would be worthwhile to briefly explore normal -- or "bell-shaped" -- curves in Excel. This is a commonly used area of statistics, and one for which Excel provides several useful functions.

One interesting thing about the normal curve is that it occurs frequently in many different settings:

• The height of each gender in a population is normally distributed.
• The measure of LDL cholesterol is normally distributed in adults.
• The width of stripes on a zebra is said to be normally distributed.
• Most measurement errors are assumed to be normally distributed.
• Many Six-Sigma calculations assume normal distribution.
• Etc.

As a final example, here's a surprising occurrence of the normal curve: Take any population, whether it's normally distributed or not. Randomly select at least 30 members from that population, measure them for some characteristic, and then find the average of those measures. That average is one data point. Return the samples, select another random sample of the same number, and find the average of their measures. Do the same again and again. The Central Limit Theorem says that those averages tend to have a normal distribution.

Normal distributions are all around us. Therefore, as painlessly as possible, let's take a closer look at how we work with them using Excel.
 

Brief Definitions

We need to get some brief definitions out of the way so that we can start to describe data using Excel functions.

From cholesterol to zebra stripes, the normal probability distribution describes the proportion of a population having a specific range of values for an attribute. Most members have amounts that are near the average; some have amounts that are farther away from the average; and some have amounts extremely distant from the average.

For example, a population could be all the stripes on all the zebras in the world. The normal curve would show the proportion of stripes that have various widths.

The standard deviation of a sample is a measure of the spread of the sample from the mean. In a normal distribution, about 68% of a sample is within one standard deviation of the mean. About 95% is within two standard deviations. And about 99.7% is within three standard deviations. The numbers in this figure mark standard deviations from the mean.

The z value is the distance between a value and the mean in terms of standard deviations. In the figure above, each number is a z value.
 

Calculating or Estimating the Standard Deviation

Several of the following functions require a value for the standard deviation. There are at least two ways to find that value.

First, if you have a sample of the data, you can estimate the standard deviation from the sample using this formula:

=STDEV(range_of_values)

Second, if you're working with rough estimates, you must take a different approach, because you don't have actual data to support your estimates.

In this case, first calculate the Range. This is the smallest likely value subtracted from the largest likely value. Let's assume that all possible values will be within that range about 95% of the time.

Remember that about 95% of a sample is within two standard deviations on each side of the mean. (This is a total of four standard deviations, of course.) Therefore, if we divide the range by four we should have the approximate standard deviation.

Merely dividing the range by four might seem to be a slipshod approach. But consider the way this calculation often is used.

Suppose you're forecasting sales for next year. You think sales will be about 1,000, but the number could be as high as 1,200 and as low as 800. With that information, you can put a normal curve around your estimated sales and begin to generate a variety of forecasts for profits and cash flow.

To emphasize, these numbers are only your best estimates. Therefore, using an estimated standard deviation doesn't seem quite as sloppy as it otherwise might.

Based on these estimates, your mean sales will be about 1,000 and your standard deviation will be about (1200 - 800) / 4 = 100. With this information, you can use the following functions to perform many of the calculations you will need in your analysis.
 

NORMDIST(x, mean, standard_dev, cumulative)

NORMDIST gives the probability that a number falls at or below a given value of a normal distribution.

• x -- The value you want to test.
   
• mean -- The average value of the distribution.
   
• standard_dev -- The standard deviation of the distribution.
   
• cumulative -- If FALSE or zero, returns the probability that x will occur; if TRUE or non-zero, returns the probability that the value will be less than or equal to x.

Example: The distribution of heights of American women aged 18 to 24 is approximately normally distributed with a mean of 65.5 inches (166.37 cm) and a standard deviation of 2.5 inches (6.35 cm). What percentage of these women is taller than 5' 8", that is, 68 inches (172.72 cm)?

The percentage of women less than or equal to 68 inches is:

=NORMDIST(68, 65.5, 2.5, TRUE)  =  84.13%

Therefore, the percentage of women taller than 68 inches is 1 - 84.13%, or approximately 15.87%. This value is represented by the shaded area in the chart above.
 

NORMSDIST(z)

NORMSDIST translates the number of standard deviations (z) into cumulative probabilities.

To illustrate:

=NORMSDIST(-1)  = 15.87%

=NORMSDIST(+1) = 84.13%

Therefore, the probability of a value being within one standard deviation of the mean is the difference between these values, or 68.27%. This range is represented by the shaded area of the chart.

NORMINV(probability, mean, standard_dev)

NORMINV is the inverse of the NORMDIST function. It calculates the x variable given a probability.

To illustrate, consider the heights of the American women used in the illustration of the NORMDIST function above. How tall would a woman need to be if she wanted to be among the tallest 75% of American women?

Using NORMINV, she would learn that she needs to be at least 63.81 inches tall, as shown by this formula:

=NORMINV(0.25, 65.5, 2.5)  = 63.81 inches

The figure shows the area represented by the 25% of the American women who are shorter than this height.
 

NORMSINV(probability)

NORMSINV is the inverse of NORMSDIST function. Given the probability that a variable is within a certain distance of the mean, it finds the z value.

To illustrate, suppose you care about the half of the sample that its closest to the mean. That is, you want the z values that mark the boundary that is 25% less than the mean and 25% more than the mean.

The following two formulas provide those boundaries of -.674 and +.674, as illustrated by the figure.

=NORMSINV(0.25)
=NORMSINV(0.75)

STANDARDIZE(x, mean, standard_dev)

STANDARDIZE returns the z value for a specified value, mean, and standard deviation.

To illustrate, in the NORMINV example above, we found that a woman would need to be at least 63.81 inches tall to avoid the bottom 25% of the population, by height. The STANDARDIZE function tells us that the z value for 63.81 inches is:

=STANDARDIZE(63.81, 65.5, 2.5)  =  -0.6745

We can check this number by using the NORMSDIST function:

=NORMSDIST(-0.6745) = 25%

That is, a z value of -.6745 has a probability of 25%.
 

How to Calculate a Random Number from a Normal Distribution

Remember that the NORMINV function returns a value given a probability:

NORMINV(probability, mean, standard_dev)

Also, remember that RAND() function returns a random number between 0 and 1. That is, RAND() generates random probabilities. Therefore, it seems logical that you could use the NORMINV function to calculate a random number from a normal distribution, using this formula:

=NORMINV(RAND(), mean, standard_dev)

However, Jerry W. Lewis -- a former Excel MVP and a professional statistician -- offers a stern comment about this approach. "Using NORMINV(RAND(),...) to generate Normal variates is totally unacceptable prior to Excel XP, and is only marginal in XP. This is because of inadequacies in NORMINV and in the tails of NORMDIST itself."

"NORMINV prior to Excel XP produced a very un-normal fraction of values around 6 million standard deviations from the mean," Jerry wrote. "This is due to inaccuracies in the implementation of NORMDIST and NORMINV. Excel XP brought those values into a less obviously wrong location, but otherwise did little to improve the situation. NORMINV in Excel 2003 is a decent implementation."

Instead, Jerry recommends the Box-Muller method described here: http://mathworld.wolfram.com/Box-MullerTransformation.html
This method, he wrote, is limited only by the inadequacies of the RAND() function prior to Excel 2003, which had unacceptable autocorrelation.

The Box-Muller approach suggests that Excel users should use this formula to calculate a random number from a normal distribution:

=SQRT(-2*LN(RAND()))*SIN(2*PI()*RAND())

The Box-Muller method is mathematically exact, Jerry writes, if implemented with a perfect uniform random number generator and infinite precision.
 

A Note About The Charts

I created all of the figures for this article in Excel. If you would like to know how I created them, check back in several weeks. I'll explain how you can create similar charts, and I'll point you to a web site that provides more about this topic.

References

I must have at least 15 statistics books gathering dust on bookshelves in the basement. Even so, these two books offered clear explanations that you might find useful:

Statistical Analysis with Excel for Dummies, by Joseph Schmuller, PhD
Excel Data Analysis for Dummies, by Stephen L. Nelson, MBA, CPA

(Email Comments)