0
$\begingroup$

I am getting familiar with SPSS, specifically the Automatic Linear Modeling functionality. I first ran SPSS, using Best Subset method which gives me the result of "best" variables. Being uncertain, I run again in Excel with the same data. Much to my surprise, the coefficients of one variable are not the same: they are the opposite of each other, having identical attributes. I am ok with the different intercepts but opposite coefficients imply very different stories about the data.

What is the meaning of this? What did I do wrong? What model should be used? Does this mean I have to check again everytime SPSS return a model from best subset functionality?

Excel result

enter image description here

This is the transformation summary.

Role (Summer_transformed) Predictor

Actions Taken (Summer_transformed) Change measurement level from continuous to ordinal

Role (Month_transformed) Predictor

Actions Taken (CK_transformed) Trim outliers

Actions Taken (Month_transformed) Trim outliers

Role (CK_transformed) Predictor

$\endgroup$
1
$\begingroup$

It seems that SPSS uses summer_transformed=1 as the reference group, while Excel uses summer_transformed=0. You can see that the difference in intercepts is also equal to the difference between these groups, so once you correct the reference, you should get exactly the same results.

Two side comments: 1) I wouldn't use values in the billions when learning a method, just to see the results easier; 2) I always recommend R instead of SPSS as your second analysis tool, if you want to understand the methods and are not afraid to get hands dirty with some coding.

$\endgroup$
  • $\begingroup$ Pt 2 can be irrelevant to the question, particularly it was not supported in the answer. SPSS has syntax/coding user language, and SPSS analysis methods documentation (algorithms description etc.) is better than R's documentation. $\endgroup$ – ttnphns Jun 21 '17 at 7:30
  • 1
    $\begingroup$ @ttnphns I personally find that R's requirement to use syntax instead of menus and dialogs (assuming that we're not talking about R-commander and such) forces students to think more of what happens under the hood. But of course, that's just a subjective recommendation, YMMV. $\endgroup$ – juod Jun 21 '17 at 8:03
  • 1
    $\begingroup$ I agree with your last comment, especially for students thinking . Adding, however a minor note that "thinking" all the time in routine job (i.e. doing only syntax, w/o menus) can be really exhausting to mind. Menus are fine gimmics for distraction and labour diversity. Smaller blood pressure after work. $\endgroup$ – ttnphns Jun 21 '17 at 10:07
  • $\begingroup$ @juod: I am not sure what would "correct the reference" entail. The transformation for the Summer variable is "Change measurement level from continuous to ordinal". In my data, Summer variable only have value of 0 and 1. [link]i.stack.imgur.com/wPhFc.png $\endgroup$ – Tam Le Jun 21 '17 at 10:24
  • 1
    $\begingroup$ @juod: I could understand that. It would appear that my value of 0 was turned to 1 and 1 to 0 in SPSS's calculation. But why would SPSS do something like that? That essentially undermine my data construction and ultimately, my hypothesis. It would also mean that I have to flip over coefficient every time I see "continuous to ordinal" transformation. It seems counter-productive but I guess SPSS does not do that without good reason. So what is that? $\endgroup$ – Tam Le Jun 22 '17 at 2:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.