Significant Levene's Test for Homogeneity

[Yughaber]

I have a dataset where I want to see if the time of the day (divided into categories of 4 hour intervals) has an effect/makes a difference in the length of stay of visitors. I thought to use one way ANOVA, however, when running the Levene's Test for homogeneity, I got a significant p-value which doesn't meet my assumption of equal variance. Any suggestions of what is the best test to use in this case? Thanks in advance! 

[PGStats]

First, check your data for outliers. Otherwise, the Wilcoxon test provided in proc NPAR1WAY is a good alternative.

[Yughaber]

take out outliers if there are any?

[PGStats · Accepted solution]

Take out outliers if they are truly aberrant (i.e. measurement errors or measurements outside the scope of your study). But don't throw them out just for the sake of running a parametric test. Use a robust test like the one I suggested instead.

[Yughaber]

I see, ok that sounds nice. How would I code for the proc NPAR1WAY?

[PGStats]

Simple. For example, does car engine size differ among car models of different origins?

proc npar1way data=sashelp.cars wilcoxon;
class origin;
var enginesize;
run;

Note, the Wilcoxon test is called Kruskal-Wallis when comparing more than 2 categories.

[Yughaber]

awesome! Thanks!!

[Yughaber]

For the median plot in NPAR1WAY, do you know how I can display how many counts/frequencies are above the median and how many are below? The plot only gives a total N and just shows a graph of above and below the median but I need to also know the frequencies above and below...

[jiltao]

You might also consider using PROC MIXED. PROC MIXED allows you to model the unequal variances. For example,

proc mixed;

   class hour;

   model y=hour;

   repeated / group=hour;

run;

This program estimates different variances in y for different groups in HOUR, and the inferences would take into account different variance estimates. 

[Yughaber]

this will only explain the variance right? I can't go further with proving any significant differences between the different time stamped visitor stays

[jiltao]

Yes, you can use the LSMEANS statement in PROC MIXED to do comparisons.