Before you put your faith in statistics, reassure yourself that you know both the value and the limitations of the techniques that you use. Statistical methods are just tools—they cannot guard you from incorrect science (invalid statistical assumptions) or bad data.
Most statistics are based on the assumption that the model is correct. To the extent that your model might not be correct, you must attenuate your credibility in the statistical reports that result from the model.
Many statistical tests do not evaluate the model in an absolute sense. Significant test statistics might be saying only that the model fits better than some reduced model, such as the mean. The model can appear to fit the data but might not describe the underlying physical model well at all.
Often the value of the statistical results is not that you believe in them directly, but rather that they provide a key to some discovery. To confirm the discovery, you might need to conduct further studies. Otherwise, you might just be sifting through the data.
For example, if you conduct enough analyses, you can find 5% significant effects in 5% of your studies, even if the factors have no predictive value. Similarly, to the extent that you use your data to shape your model (instead of testing the correct model for the data), you are corrupting the significance levels in your report. The random error then influences your model selection and leads you to believe that your model is better than it really is.
There are a variety of techniques and patterns to assess the validity of the model:
• Model validity can be checked against a saturated version of the factors with Lack of Fit tests. The Fit Model platform presents these tests automatically if your data contain replicated x values in a model that is not saturated.
• You can check the distribution assumptions for a continuous response by looking at plots of residuals and studentized residuals from the Fit Model platform. Or, use the Save commands in the platform pop-up menu to save the residuals in data table columns. Then use the Analyze > Distribution on these columns to look at a histogram with its normal curve and the normal quantile plot. The residuals are not quite independent, but you can informally identify severely nonnormal distributions.
• The best all-around diagnostic tool for continuous responses is the leverage plot because it shows the influence of each point on each hypothesis test. If you suspect that there is a mistaken value in your data, this plot helps determine whether a statistical test is heavily influenced by a single point.
• It is a good idea to scan your data for outlying values and examine them to see whether they are valid observations. You can spot univariate outliers in the Distribution platform reports and plots. Bivariate outliers appear in Fit Y by X scatterplots and in the Multivariate scatterplot matrix. You can see trivariate outliers in a three-dimensional plot produced by the Graph > Scatterplot 3D. Higher dimensional outliers can be found with Principal Components or Scatterplot 3D, and with Mahalanobis and jack-knifed distances computed and plotted in the Multivariate platform.
The statistical literature describes special nonparametric and robust methods, but JMP implements only a few of them at this time. These methods require fewer distributional assumptions (nonparametric), and then are more resistant to contamination (robust). However, they are less conducive to a general methodological approach, and the small sample probabilities on the test statistics can be time consuming to compute.
If you are interested in linear rank tests and need only normal large sample significance approximations, you can analyze the ranks of your data to perform the equivalent of a Wilcoxon rank-sum or Kruskal-Wallis one-way test.
If you are uncertain that a continuous response adequately meets normality assumptions, you can change the modeling type from continuous to ordinal and then analyze safely. However, this approach sacrifices some richness in the presentations and some statistical power as well.