molecalculator

Moles & Molar Mass

How to Check for Normality in JMP

Guide 01-2915 · Updated · Step by step

Checking normality in JMP helps you see whether a continuous variable is reasonably consistent with a normal distribution, an assumption used by some statistical methods. You will examine a histogram, a normal quantile plot, and a formal goodness-of-fit test; the plots and the study context matter more than any single p-value.

What You’ll Need

  • A JMP data table with the observations in one column. Use a continuous measurement, such as height, time, or concentration.
  • A numeric column with enough valid observations to make a distribution plot useful. Check that missing values and special codes are identified correctly.
  • JMP installed. Menu names can vary slightly across JMP versions, but the Distribution platform is the standard starting point.

Step-by-Step

  1. Check the data column. Open your data table and confirm that the variable is stored as numeric and has the appropriate modeling type, usually Continuous. In the column information, verify that text entries, blanks, or codes such as 999 have not been mistaken for valid measurements. Do not remove unusual values just because they look inconvenient; first establish whether they are data errors or genuine observations.
  2. Open the Distribution platform. Select Analyze > Distribution from the main menu. In the launch window, select the numeric variable in the column list and click Y, Columns to assign it. If you want to inspect separate groups, you can also assign a categorical variable to By; JMP will create a separate distribution report for each group.
  3. Run the report. Click OK. JMP displays a distribution report that typically includes a histogram, summary statistics, and a box plot. For grouped data, review each group rather than relying on a pooled plot, because groups with different centers or spreads can make the combined distribution look nonnormal.
  4. Add a normal quantile plot. Click the red triangle beside the variable name at the top of its report. In the menu, choose Normal Quantile Plot (the exact wording or menu location may differ by release). The plot compares ordered observations with the values expected under a normal distribution. Points that follow a roughly straight line support approximate normality; systematic curves, an S-shape, or isolated points far from the line suggest departures.
  5. Request a normality test. Open the same red-triangle menu and select Continuous Fit if needed, then choose the normal distribution option or its fitting submenu. In many JMP versions, the fitted normal report has a red-triangle menu containing Goodness of Fit or a normality test. Select the available test to display the test statistic and p-value. If the menu differs, use JMP Help for your version and search for “Distribution goodness of fit” or “normal quantile plot.”
  6. Read the p-value correctly. The test’s null hypothesis is that the observations follow the specified normal distribution. With a chosen significance level, commonly 0.05, a p-value below that level is evidence against normality. A p-value at or above it means the test did not detect a departure; it does not prove the data are normal. Record the test name, sample size, and p-value with your analysis.
  7. Judge practical importance. Look at the histogram and quantile plot alongside the test. Check for skewness, heavy tails, multiple peaks, and influential outliers. Consider whether the analysis requires the raw variable to be normal or instead requires model residuals to be approximately normal. For regression or ANOVA, assess residuals from the fitted model rather than treating normality of the response by itself as the assumption.
  8. Decide what to do next. If the plots show only a mild departure and your method is reasonably robust for your sample size and design, you may proceed while reporting the check. For a serious departure, investigate data quality, group structure, and outliers; consider a suitable transformation or a method that does not rely on normality. Recheck the assumptions after making any change, and explain the rationale rather than deleting observations solely to obtain a larger p-value.

Common Mistakes

  • Treating a nonsignificant test as proof. A normality test can miss departures, especially with small samples. It may also flag trivial departures with very large samples.
  • Using the wrong variable. Many model-based methods assume normal errors or residuals, not that every original measurement is normally distributed. Check the assumption for the analysis you plan to run.
  • Ignoring dependence or groups. A normality test does not establish independence. Repeated measures, time trends, or mixed populations need separate consideration.
  • Removing outliers automatically. First verify entries and measurement records. A legitimate extreme value may be scientifically important and should not be removed merely to improve a plot or test result.
  • Assuming all tests are identical. JMP’s available tests and report labels can differ by version and fitting options. State which test you used instead of reporting only “the normality test.”

Tips

  • Use at least two views: a histogram for overall shape and a normal quantile plot for departures in the tails.
  • Check sample size and the sampling process. A plot can look irregular with few observations, while a large sample can reveal deviations that have little practical impact.
  • When data are grouped, inspect each group and the model residuals as appropriate. A pooled distribution can hide meaningful differences.
  • Keep a record of JMP’s version, selected test, significance level, and any exclusions or transformations so that the check can be reproduced.

Troubleshooting

  • The variable does not appear in the launch window: Confirm that the column is numeric and not set as Nominal or Ordinal when a continuous variable is intended. Correct the column’s data type or modeling type in the table, then reopen Distribution.
  • The histogram or test looks wrong: Check for missing values, coded missing values, duplicate records, mixed units, and data-entry errors. Confirm the column contains one comparable measurement rather than several combined variables.
  • You cannot find the normality test: Open the report’s red-triangle menus and look for the continuous fit and goodness-of-fit options. Some choices appear only after fitting a distribution. Menu labels vary by release, so consult the built-in help for your installed version.
  • The test rejects normality but the plot looks reasonable: With a large sample, small deviations can be statistically detectable. Examine the size and location of departures and decide whether they matter for the intended analysis.
  • The plot looks nonnormal but the test does not reject: A small sample may not provide enough evidence for the test to detect a departure. Review the data and use methods appropriate to the design; do not treat the p-value as a guarantee.

Frequently Asked Questions

What is the quickest way to check normality in JMP?
Use Analyze > Distribution, assign the numeric column, and review its histogram and normal quantile plot. Add a goodness-of-fit test from the report’s red-triangle menu when available.

What does a straight normal quantile plot mean?
Points close to a straight line are consistent with an approximately normal distribution. Look for systematic curvature and tail departures, not just a few small deviations.

What p-value indicates normality?
There is no p-value that proves normality. A low p-value, often below a preselected 0.05 threshold, is evidence against the test’s normality null hypothesis; a higher value means insufficient evidence to reject it.

Should I test the raw data or residuals?
For methods whose assumption concerns model errors, inspect the model residuals. The raw response itself does not necessarily need to be normally distributed.

Can I use a normality test with a small sample?
You can, but its power may be limited. Pair the test with plots and knowledge of how the data were collected, and avoid drawing a firm conclusion from the p-value alone.

Does JMP have the same normality menu in every version?
No. Report menus and test availability can vary by version and platform settings. Use the Distribution report’s red-triangle menus or JMP Help to locate the corresponding option.

Related

General steps for current mainstream versions; menus can differ by device and version. See the disclaimer or report an error.

More in this section

The whole Moles & Molar Mass section →