2

I am performing a one-way ANOVA in R with the following data:

Cu Day CC
Cu1 49 30934500
Cu1 49 26860125
Cu1 49 46524750
Cu10 49 15272561
Cu10 49 31601659
Cu10 49 17627634
Cu100 49 3718127
Cu100 49 4941416
Cu100 49 6230801

The ANOVA aims to determine if CC changes (on average) depending on Cu (three levels: Cu1, Cu10, and Cu100) or, in other words, to check if Cu exerts an effect on CC. The code I used was:

    analisis_varianza <- aov(CC~Cu, data=dia_49)
    summary(analisis_varianza)

And, to check the homoscedasticity assumption:

   leveneTest(CC~Cu, data=dia_49)

The problem is that Levene's test suggests that variances are homogeneous (p=0.5424), while the residuals vs fitted plot for the ANOVA clearly indicates that the variance is not homogenous:

enter image description here

On which should I rely on and why?

mkt
  • 18,245
  • 11
  • 73
  • 172
Sergio M.
  • 21

1 Answers1

1

In general, it's a bad idea to rely on tests of normality for taking decisions. In this case, the Levene's test p-value is probably high because you have a very small sample size. On the other hand, if you have very large sample sizes, the test will show significant deviations from normality even when it's not a problem for inference.

In short, graphs are much more informative than tests.

Here's a very useful thread: Is normality testing 'essentially useless'?

mkt
  • 18,245
  • 11
  • 73
  • 172