I have tried a few things to try to normalize my data but none worked (LN, SqRt, etc). Please, help. Here are the two DV (Income and éducation). I need to use them as continues variables. SPSS.
1) What is your monthly earned income before taxes?
□ 0-300 KW □ 301-600 KW □ 601-900 KW
□ 901-1200 KW □ 1201-1500 KW □ 1501 KW+
2) What is your highest level of education achieved?
□ Have not attended school □ Completed some school □ Finished 7 grades □ Finished 9 grades □ Finished 12
grades □ Finished tertiary education and beyond
Thank you!!
You could recode these categories to Likert scales:
e.g., 1 = 0-300KW, 2 = 301-600KW, 3 = 601-900KW, 4 = 901-1200KW, 5 = 1201-1500KW and 6 = >1501KW
and for education levels:
e.g., 1 = not attended - - 6 = finished tertiary and beyond.
But, why do you want to use these as continuous variables? If you would want to assess the association between these categories and for example Gender, then you could run a Chi-square test. Idem for the association between Income and Education.
Jos Feys its already on the Likert scale. I need to see the relationship between parental involvement scale (average scale or factors) with Income and education.
Dear Yuliya,
why do you want to "normalize" these variables? If they are skewed: No need to transform them as there is no normality assumption regarding the DV in regression, only with regard to the residuals of the regression. Hence, you could test that. I do not know whether SPSS allows plotting qqplots (google what that is) or outlier statistics (e.g., Cooks distance). If it does not, switch to R. If you this, check the tidyverse approach by H. Wickham ( https://rviews.rstudio.com/2017/06/08/what-is-the-tidyverse/ ) which makes using R so much easier.
For a bit deeper look, see
Wickham, H., & Grolemund, G. (2016). R for data science: Import, tidy, transform, visualize, and model data. Beijing: O'Reilly Media, Inc.
On the other hand, it is plausible that especially the income measure is a coarsed version of an underlying gamma-distributed variable. You could thus compare the results of a OLS regression with a GLM model (generalized linear model) with a gamma error distribution. Sounds complicated but in R quite easy.
Ng, V. K. Y., & Cribbie, R. A. (2017). Using the gamma generalized linear model for modeling continuous, skewed and heteroscedastic outcomes in psychology. Current Psychology, 36(2), 225-235. doi:10.1007/s12144-015-9404-0
Barber, J., & Thompson, S. (2004). Multiple regression of cost data: use of generalised linear models. Journal of Health Services Research & Policy, 9(4), 197-204. doi:10.1258/1355819042250249
From a more rigid point of view, your measures (esp. education) consist of rough categories with unknown intervals, hence, applying an ordinal regression would be a further option.
HTH
Holger
Holger Steinmetz thank you for the answer. Residuals are not normally distributed. Most people I interviewed had low income or high. Hence, the middle is « empty ». I would like to see if I can somehow transform the data to be normally distributed. Education is a bit closer to the normal distribution. Any suggestions how to transform it? Thx
Yuliya Lynch following Holger's reply, I think it is better to try to accommodate the distributional nature of your variables. That way you can address the specifics of your dataset and may get more meaningful results.
- Badges
- Science topic
- Similar topics
- Teacher Education
- School