molecalculator

Moles & Molar Mass

How to Test Normality in JMP: A Step-by-Step Guide

Guide 01-2916 · Updated · Step by step

Testing normality in JMP helps you judge whether a continuous variable is reasonably consistent with a normal distribution, an assumption used by some statistical methods. You will need a column of numeric observations and can use JMP’s distribution plots and normality tests together; a test result is evidence to interpret, not an automatic pass-or-fail verdict.

What You’ll Need

  • A JMP data table containing the observations you want to assess.
  • A continuous numeric column, such as a measurement or residual, with missing or invalid entries identified.
  • JMP with the Distribution platform available. Menu wording can vary slightly by JMP version.

Step-by-Step

  1. Choose the right values to test. Decide what normality assumption applies to your analysis. For many models, the relevant assumption concerns model residuals, not necessarily the original response column. If you are checking a sample as a preliminary assessment, select the measured variable. Do not combine distinct groups or conditions unless the analysis treats them as one population.
  2. Check the data column. Open the JMP data table and confirm that each row is one observation and the target column is numeric and continuous. Look for text entries, special missing values, data-entry errors, and values recorded in inconsistent units. Decide how to handle missing observations using the rules for your analysis; do not silently replace them with zero.
  3. Open the Distribution platform. From the top menu, choose Analyze > Distribution. In the launch window, select your numeric column and move it into Y, Columns. If you need separate assessments by category, place the grouping column in By or use the appropriate grouping role for your JMP version. Click OK.
  4. Show the normal quantile plot and tests. In the report, find the red triangle menu beside the variable’s heading or distribution report. Choose Continuous Fit > Normal if a normal curve is not already displayed. Open the relevant red triangle menu and select Normal Quantile Plot or Normality Test, depending on your version and report layout. The quantile plot compares observed values with values expected under normality; the test report commonly includes the Shapiro–Wilk test and may show other tests depending on sample size and version.
  5. Read the test output. Find the test statistic and its p-value. The null hypothesis for a normality test is that the data are consistent with a normal distribution. A small p-value, often compared with a preselected level such as 0.05, is evidence against that null. A p-value above the threshold does not prove normality; it means the test did not find sufficiently strong evidence against it.
  6. Inspect the plots and summary. Review the normal quantile plot for systematic curvature, pronounced tail departures, or isolated points. Check the histogram and box plot for skew, multiple peaks, and outliers. Consider sample size: a small sample may not reveal meaningful departures, while a very large sample can flag deviations that have little practical effect.
  7. Decide what the result means for your analysis. Consider the method’s sensitivity to non-normality, sample size, independence, and equal-variance assumptions. If you are checking regression or analysis-of-variance assumptions, inspect residuals and other diagnostics as well. If departures are important, discuss a justified transformation, robust method, or suitable alternative with your statistical plan.
  8. Save or report the result. Keep the data context, number of usable observations, test name, statistic, p-value, and relevant plots with your analysis. Use JMP’s report controls to save or copy the output. State that the assessment did not find evidence against normality rather than claiming that normality has been proven.

Common Mistakes

  • Testing the wrong quantity: A model may require approximately normal residuals, not a normally distributed raw response. Save the model’s residuals and assess those when that is the stated assumption.
  • Relying only on a p-value: Test power depends on sample size. Always check plots and consider whether departures matter for the intended analysis.
  • Mixing unlike observations: Combining separate populations, batches, or groups can create a non-normal overall shape even when each group is reasonable on its own. Assess groups separately when the design calls for it.
  • Deleting outliers just to improve the result: Investigate unusual observations against source records and study rules. Correct verified errors, but do not remove valid data solely because they affect the test.
  • Confusing the null hypothesis: A significant result is evidence against normality; a non-significant result is not proof that the data are normal.

Tips

  • Choose the significance threshold before looking at results, and report it alongside the p-value.
  • Use the normal quantile plot as a practical visual check. Points close to a straight line support, but do not establish, an adequate normal approximation.
  • For grouped data, review each relevant group and consider whether the sample size within each group is sufficient for a meaningful test.
  • Keep the test tied to the analysis. Normality is only one possible assumption; independence, correct model form, and constant variance may be equally or more important.

Troubleshooting

  • The Distribution platform will not run: Check that the column is assigned to Y, Columns, is numeric, and contains usable values. If it is stored as character data, correct or convert it only after confirming the entries represent numbers.
  • No normality test appears: Open the red triangle menus in the Distribution report and look under the continuous-fit or distribution options. Some versions show the test after fitting a normal distribution. Menu names and available tests differ between releases.
  • The plot or test looks unexpected: Confirm units, decimal placement, missing-value handling, and whether a filter or data subset is active. Check whether the observations are independent and whether multiple groups were pooled.
  • The p-value conflicts with the plot: This can happen because tests and visual checks respond differently to sample size and types of departure. Report both, inspect the tails and outliers, and interpret them in the context of the planned method rather than choosing whichever conclusion is convenient.
  • There are too few observations: JMP may still provide a plot or test, but evidence can be weak with a small sample. Collect more data if appropriate, and avoid treating a non-significant result as confirmation of normality.

Frequently Asked Questions

Which normality test does JMP use?
JMP commonly reports the Shapiro–Wilk test in its normality output, with available tests depending on the JMP version and context. Check the test name shown in your report.

What does a p-value below 0.05 mean?
If 0.05 was your preselected threshold, it indicates evidence against the normality null hypothesis at that threshold. It does not tell you how large or practically important the departure is.

Does a p-value above 0.05 mean my data are normal?
No. It means the test did not detect enough evidence against normality. Limited sample size or low test power may also explain the result.

Should I test the raw data or residuals?
Follow the assumption of the analysis. Regression and many model-based procedures concern residuals; a general distribution check may concern the raw variable.

Can I test normality separately by group?
Yes. Use a grouping role or assess subsets separately when groups represent distinct populations or model conditions. Interpret results with each group’s sample size in mind.

What should I do if the data are not normal?
First check for errors, influential observations, and mixed populations. Then consider whether the planned method is robust or whether a justified transformation or alternative method is suitable.

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 →