@Francois: You should demonstrate the superiority of GLM to the proposed optimal data analysis approach for this particular analytic problem. Which of the two approaches provides a more accurate answer is not a matter of personal opinion, but rather an objective reality. Show me the data that confirm your opinion that GLM is more accurate than ODA in this particular application. 

This empirical challenge contains no "delusion of grandeur" and no "missionary zeal," merely a reminder that science is a matter of fact and not a matter of opinion. If you claim that GLM is a superior approach to ODA in solving this particular quantitative problem, then provide evidence to support that claim. Otherwise, all you are offering is a personal opinion that lacks supporting evidence.  Show me the data.

There is strong evidence to support the superiority of ODA to GLM. What evidence is there that GLM is superior to ODA? Why should researchers use GLM, if ODA provides more accurate answers? Delusions of accuracy?

just to clarify it: your original variables had 3/9 or 4/10 categories? because it they had 3/9 categories and you entered all 3/9 dummy variables it will not work. you can only enter 2 dummy variables if the original variable has 3 categories.

is this your Problem? if not, what is it? another question: what's your 9/10 category variable and why recode it into dummy variables?

My original variables had 3 and 9 categories. The variable I recoded into 9 dummy variables is number of children (zero through to eight or more). I do understand that I must have 1 less dummy variable than the total of original categories in the variable.

My question is whether or not unrelated dummy variables (whether that be 3/9 or 2/8) can be included in the same multiple regression using the Enter method? It seems incorrect to have 10+ predictor variables in a regression analysis.

Thank you.

it's not incorrect, but of course you should have good/plausible theoretical reasons for entering that many variables. and you need adequate number of cases.

I - just my personal opinion and expercience which may not be statistically correct - would not recode the "number of children" variable into dummy variables but enter it like it is (if there is a linear relationship if you do linear Regression). you also can recode it into lesser dummy variables (if - again - it's plausible), e.g. 0, 1-2, 3-4 Kids etc.

I have 1400+ cases so a high number of predictors is not a problem (if my maths is correct).

I was also thinking that number of children could be entered as is. I'm going to try running an analysis and see how it turns out.

Thanks a lot.

I've seen similar approaches with credit rating, when you are trying to assess different effects along the different rating scales. In this case, are you hypothesising different effects over the dependent variable for different change in the number of kids? (i.e. increasing 1 kid when already having 3 would have a different effect that if the subject had none)

Is number of children an ordinal variable? It seems like ratio to me (unless coded in an ordinal way such as 0, 1-2, 3-5, 5+ etc.).

it could be ordinal since the highest category is 8+.

but the question is, if there are cases in this last category. if not, then it's a ratio scale from 0 to 7.

There are cases in the 8+ category but these will be removed due to a high number of outliers within.

Do you have the raw data? Those values probably aren't outliers in any meaningful sense - some families have more than seven children. In a multiple regression what matters is the distribution of the residuals not the predictors.

Regression Analysis and ANOVA are both examples of the general linear model (GLM). The traditional ANOVA model uses exclusively categorical predictors and the traditional Regression model uses scale predictors. If you have categorical and scale predictors, you are actually in a General Linear Model situation. For your problem I would rather use the General Linear Model routine in SPSS. It saves you having to create binary variables, because it is done automatically in GLM. You enter the categorical variables as "FIXED FACTORS" and the scale  variables as "COVARIATES". You may also include interaction terms in the model. You have to specifically ask for the parameter estimates under OPTIONS in order to obtain the regression coefficients. All else being the same, you should get exactly the same model predictions under Regression and GLM.

Illusions of grandeur?

Sorry, but at my age I am not susceptible to the missionary zeal of a new religion.

@Francois: You should demonstrate the superiority of GLM to the proposed optimal data analysis approach for this particular analytic problem. Which of the two approaches provides a more accurate answer is not a matter of personal opinion, but rather an objective reality. Show me the data that confirm your opinion that GLM is more accurate than ODA in this particular application. 

This empirical challenge contains no "delusion of grandeur" and no "missionary zeal," merely a reminder that science is a matter of fact and not a matter of opinion. If you claim that GLM is a superior approach to ODA in solving this particular quantitative problem, then provide evidence to support that claim. Otherwise, all you are offering is a personal opinion that lacks supporting evidence.  Show me the data.

There is strong evidence to support the superiority of ODA to GLM. What evidence is there that GLM is superior to ODA? Why should researchers use GLM, if ODA provides more accurate answers? Delusions of accuracy?

In my opinion, flexibility to catch dynamic and stochastic behavior of the data is key to any statistical modeling. OLS assumptions are constantly violated in the real data world and linear model does not catch real characteristics of the data.

I firmly believe that model should follow data rather than the opposite. I can sculpture most of the data to fit nicely with a model, but that does not realistically reflect the objectives does it? Categorizing is overall better than trimming outliers.