Getting the coefficient for main effects in an ordinal variable - least-sq regressionHi, I have an ordinal variable (health literacy, with 5 levels). I am testing it against a continuous dependent variable using a least square regression. The output gives me betas for each level of my ordinal, but how do I get the main effect across all levels of ordinal, like SPSS gives you a main coefficient over all levels of the ordinal variable?
macOSRe: Getting the coefficient for main effects in an ordinal variable - least-sq regressionThe effect coding used in JMP results in the sum of the effects of all the levels equal to zero, if I understand your question. You can get the overall test for the ordinal term below the parameter estimates.
Re: Getting the coefficient for main effects in an ordinal variable - least-sq regressionThanks, is there a coefficient (beta) associated with the overall term "Hispanic health literacy"? When I run the stats in SPSS it gives me a single beta coefficient for Hispanic health literacy (instead of for its 5 levels, which I can see in .jmp under expanded estimates).
Any advice on how to get that single beta for the overall main effect? Like here in the screenshot, can I get a single beta instead of one associated with each of the levels of the variable)?
Re: Getting the coefficient for main effects in an ordinal variable - least-sq regressionCan you show me the same output from SPSS?
What parameterization does SPSS use for your model? Here is what JMP is doing (nominal factors and ordinal factors), so it doesn't make sense (to me) to have a single coefficient. I looked for examples of SPSS output for models with categorical factors. None of them show a single parameter for such predictors.
Re: Getting the coefficient for main effects in an ordinal variable - least-sq regressionI wonder if SPSS is treating the ordinal variable as a continuous one? If so, you can get the same behavior from JMP by right-clicking on the health literacy variable under the Columns' panel of the data table and selecting "Continuous." Then, repeat the step to fit your least squares regression. You should make sure that you want to make the assumption that the data are continuous rather than treating them as ordinal.
Best,
~Laura
Re: Getting the coefficient for main effects in an ordinal variable - least-sq regressionThe factor levels in the picture of his analysis are not numeric, so they cannot be treated as continuous. They would have to first be recoded to a number (data type is numeric) before making it a continuous factor (model type is continuous).
Re: Getting the coefficient for main effects in an ordinal variable - least-sq regressionImporting data from SPSS often leads to value labels being created, such as:
If this is the case, just changing to "Continuous" will be enough--but if there aren't any value labels, then recoding first to a number will be needed.
Re: Getting the coefficient for main effects in an ordinal variable - least-sq regressionOK got it, SPSS does treat it as a continuous and gives me a single coefficient for health literacy. I recoded the levels as 1,2,3,4,5 and .jpm did the same - it's a different question if I "should" use it as a continuous variable... but that helps! Got it!! Thanks you all!
Re: Getting the coefficient for main effects in an ordinal variable - least-sq regressionYou should treat the factor as continuous if the ordered levels make sense in your interpretation of the problem. This method assumes that the difference between all the levels is the same. That is, it is a uniform change in the interval between the first and the second, the second and the third, and so on.. The parameter that is estimated is the linear slope, or change in the response per unit change in the factor over the entire domain of X. This assumption also affects the interpretation of the standard error of the estimate and the t-test for the slope. If you don't believe and can't justify such uniform intervals between levels, than even though you can estimate this slope, you probably shouldn't.
