You can use the UNIANOVA procedure (Analyze > General Linear Model > Univariate in the GUI) rather than the REGRESSION procedure.  Include categorical predictor variables as fixed factors and continuous/scaled predictors as covariates.  There is an option to display parameters, which will give you a table of coefficients similar to what you get via REGRESSION. 

You can use the UNIANOVA procedure (Analyze > General Linear Model > Univariate in the GUI) rather than the REGRESSION procedure.  Include categorical predictor variables as fixed factors and continuous/scaled predictors as covariates.  There is an option to display parameters, which will give you a table of coefficients similar to what you get via REGRESSION. 

Hi,

Kindly see the following link:

http://www.psychstat.missouristate.edu/multibook/mlt08m.html

 You try the nominal logistic regression and regression with  making dummy variable.

look these links

https://www.youtube.com/watch?v=ywGaWScf8RM

https://www.youtube.com/watch?v=R1NqbkaWO5Y

https://www.youtube.com/watch?v=ZyruUhomEnQ

Hi @Bruce, would this work for multivariate GLM too? If yes, is there a limit to how many variables (categorical and continuous) the multivariate GLM requires?

Thanks

Hello Mohammed. Yes, it will work for multivariate GLM too. See the documentation for info on limitations.

• https://www.ibm.com/support/knowledgecenter/en/SSLVMB_25.0.0/statistics_reference_project_ddita/spss/advanced/syn_glm_multivariate_overview.html

A quick glance suggests that the limitations you run into will be hardware related. HTH.

Dear colleagues, to continue the conversation, Is it permissible to have multiple dummies in a linear regression. I am talking about 22 dummies. I am looking at the effects of federal gov departments n22 (as one variable among others) on the effective government operation measured as index. The 22 units of dummy variables scares me. What problems might occur?

Hello Tamara. I don't see anything wrong with it in principle, provided your sample size is large enough to avoid overfitting. (See Mike Babyak's nice article on overfitting for more info about that--link below.) If one has a categorical explanatory variable with a large number of levels, and if those levels represent a random sample from some population of levels, then one could estimate a random intercepts (or random coefficients) model instead of using OLS linear regression. But you have 22 federal government departments, so I imagine you want to treat your variable as fixed rather than random. Is there some sensible (a priori) way to devise a smaller number of groups of departments? You could carry out some contrasts based on those groupings. HTH.

https://www.cs.vu.nl/~eliens/sg/local/theory/overfitting.pdf

Wow, Bruce, thank you very much for an expedient answer. I expected to wait for a couple of weeks for an answer. I am impressed. I will download and read the material.

These 22 are nominal, and they define some other employees' characteristics. I am not aware about the random intercepts model. But I actually grouped the depts by one characteristic - the level of employees tenure to get three groups. it helped to get OLS with good coefficients. The problem that I see is that the Durbin-Wtson is extremely low. multicolinearity shows as 1,000 and 1,000. what it means and does it matter for such models?

Bruce, I have created three groups of Fed Depts on the basis of recruitment mode of senior executives, then I dummy coded all three and run an OLS. All looks beautiful except for Durbin-Watson, which is low at .117. Is it because I am using three dummies as IVs of the the same variable (22 Fed depts)? What could be done? The three types of recruitment are my main explanatory variables.) . The Hypothesis is that meritorious recruitment improves Gov effectiveness (DV) which is interval. Am I missing some important limitation?

Hi Tamara. What I had in mind was retaining your original Dept variable, and using a series of /CONTRAST subcommands to make the contrasts of interest. Assuming you are using UNIANOVA (or GLM), they would be SPECIAL contrasts. You can see at least one example here:

• https://stats.idre.ucla.edu/spss/faq/how-can-i-do-anova-contrasts-in-spss/

HTH.

Hi Bruce, thank you very much. I will try doing the Contrasting sub commands in GLM. I was using OLS, and realized that my DVs are not normally distributed. this is my major problem. the data was collected from individual official webpages of Senior gov employees. They are 381 out of 424, those who had data available. The data on Departments effectiveness was selected from government sites. these are all kind of indices. Only later I realized that each measure of effectiveness is related to the Dep, not to individual employees, whom I study. therefore, the senior executives, who come from the same Dep. have the same measure of effectiveness. The indices are grouped rather than individualized. this grouping makes the DV abnormally distributed. Anova and other instruments do not work with grouped numerical data. I am at the dead end now. I do not know what else could be done besides descriptive statistics :(

-> hierarchical model with "Dep" as random effect?

Thanks to all for advice. I am working on transforming the effectiveness data, which looks like chunks...

If it works for my data, I will let know.

The model did not work. I guess that the log_transformation of the DV did not remove the internal problem of autocorrelation. In regression, the Durbin-watson is VERY low.

i want to ask about the appropriate model for the prediction of 4 type of learning styles with the help of background variables of the respondents

In my case the dependent as well as the predictor variables are also categorical. So what model of regression i should use and if someone please send me good reading material on how to interpret the values.@ Bruce Weaver