I'm going through the assumptions on an F-test and I want to make sure I've gone through each item so that my results are valid.

  1. Data is normally distributed;
  2. The samples are independent from one another, and;
  3. The population standard deviations of the groups are homoscedastic.

I'm using the Iris dataset and looking performing an F-test on the "sepal length (cm)" for the three different groups (setosa, versicolor, and virginica). And, for each group N=50.

enter image description here

I've verified normality for each group with Shapiro-Wilk, as none of the null hypothesis were rejected:

setosa: (0.9776989221572876, 0.4595281183719635)
versicolor: (0.9778355956077576, 0.46473264694213867)
virginica: (0.9711798429489136, 0.25832483172416687)

However when I test for homoscedasticity with Bartlett and Levene they both fail, which leads me to believe I cannot perform an F-test.

BartlettResult(statistic=16.005701874401502, pvalue=0.0003345076070163035)


LeveneResult(statistic=6.35272002048269, pvalue=0.0022585277836218586)

Does this mean that running a one-way F-test for an ANOVA will provide an incorrect analysis? I've been reading that in some cases yes it will and in others it will not.

Any guidance on where to go with the analysis is appreciated.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.