Dear ShareStats colleagues,
I saw that in the item bank a couple of items focus on Levene's test. I would suggest to remove these, or at least to add some more nuanced explanations.
In the textbook (Agresti et al.) I use at UvA it is explicitly stated not to use this test:
In Practice F Test for Comparing Standard Deviations Is Not Robust
The F test for comparing standard deviations of two populations performs poorly if the populations are not close to normal. Consequently, statisticians do not recommend it for general use. If the data show evidence of a potentially large difference in standard deviations, with one of the sample standard deviations being at least double the other, it is better to use the two-sample t-inference methods that do not have this extra assumption.
On the same page of the book a nuanced statement is given:
Recall
For large sample sizes, the normality assumption is less important because the t test is robust against violations of that assumption.
Also: Checking assumptions using tests leads to a paradoxical situation: For large $N$, most statistics are robust, but due to the enormous power the tests will give statistically significant results all the same; For small $N$, these statistics are highly sensitive to differences in variance, but due to the low power, the statistical tests will not give significant results. In short such tests are useless.
Best!
Niels
Dear ShareStats colleagues,
I saw that in the item bank a couple of items focus on Levene's test. I would suggest to remove these, or at least to add some more nuanced explanations.
In the textbook (Agresti et al.) I use at UvA it is explicitly stated not to use this test:
In Practice F Test for Comparing Standard Deviations Is Not Robust
On the same page of the book a nuanced statement is given:
Recall
Also: Checking assumptions using tests leads to a paradoxical situation: For large$N$ , most statistics are robust, but due to the enormous power the tests will give statistically significant results all the same; For small $N$ , these statistics are highly sensitive to differences in variance, but due to the low power, the statistical tests will not give significant results. In short such tests are useless.
Best!
Niels