What is the best method of checking normality through SPSS before conducting a regression?
Normality can be checked with a goodness of fit test, e.g., the Kolmogorov-Smirnov test. When the data is not normally distributed a non-linear transformation (e.g., log-transformation) might fix this issue.
Dear Hanif,
I highly recommend you to use Kolmogorov-Smirnov or Shapiro-Wilk. Use first one if your sample size is bigger than 30 and use second test if your sample size is under 30.
Please remember that the most important check of normality is on the residuals after fitting the model. Best wishes, David Booth
Analysis -- descriptive statistics -- Explore -- trans the variable to the dependent list - Plot - Normality plot s with test
Analyze - Descriptive stats - Frequency:
Statistics: check Skewness (should be near 0) // Kurtosis (should be near 3)
But I suggest you using Stata or EViews. Both will provide easy answer to your question. Hanif Qureshi
Hanif Qureshi, et.al. -
I think that the confusion many people have with normality and regression is that it is best if the Yi are close to "normally" distributed, but that refers to the conditional Y given the ith case, or
Y|predicted-y, where
Y = predicted-y + epsilon.
Actually, in addition, though it's not convenient for the Gauss-Markov Theorem, the sigma in each i-case often becomes larger with larger predicted-y, naturally. (See https://www.researchgate.net/publication/320853387_Essential_Heteroscedasticity, bounded as justified by Ken Brewer.)
The unconditional distribution of y-values, and all xs as well, can be anything. With establishment survey data, for example, the data are generally highly skewed. The residuals can still be (close to) "normally" distributed, though the variance increases with larger predictions. Weighted least squares (WLS) regression is then used. Note that OLS regression is just a special case of WLS regression, with often unrealistically equal weights. (See "When Would Heteroscedasticity in Regression Occur?" Preprint, June 2021, J. Knaub, https://www.researchgate.net/publication/352134279_When_Would_Heteroscedasticity_in_Regression_Occur.)
Cheers - Jim
Your data are not normally distributed. As Tukey (1986) says, all assumptions are wrong, and this includes the normality assumption. Thus, the K-S test is just checking if your sample size is large enough for the non-normality to be detected. Are you more interested in deciding how much the deviations from the assumptions may affect the conclusions or looking for approaches that are less influenced by these deviations.