I am using binary logistic regression. My dependent variable is referred or not referred. I have set a reference category for location of subjects. I have chosen to use the largest site with the most number of subjects as my reference group which is site 6. My question is when I look at my output, the sites are listed as numbers, site 6 is not listed, but I coded it as the reference group, so is that site on this output? the values listed in the variable view do not translate over to the output i.e. the cities. Is there a way to get the cities listed when performing the analysis?
site (Hollywood)
site 1 (Davis)
site 2 (Irvine)
site 3 (los Angeles)
site 4 (San Diego)
site 5 (san Francisco)
can I assume that site is site 6 or Hollywood?
If I reversed the decision and decided to use site 1 (DAvis) as my reference group, would I then assume that Hollywood is now coded as site 5 and and san Francisco is now 4, San Diego is now 3 etc?
Finally is there anyway to get the labeling of the sites using language vs. numbers into the output? I know I can go in and change the output manually, but looking for a way to code it into the output when setting reference groups as it is very confusing to look at what site I am looking at with just the numbers. Thanks in advance for your assistance.
-
$\begingroup$ The meaning of the so called reference category (and hence the interpretation of the regression coefficient values) is dependent on the type of contrast coding scheme. Such as whether it were indicator (dummy) coding or deviance coding. $\endgroup$– ttnphnsCommented Apr 22, 2015 at 7:23
-
$\begingroup$ ...e.g. see here. $\endgroup$– ttnphnsCommented Nov 23, 2016 at 15:42
1 Answer
Site 6 is the intercept. In effect, your model states that the log odds of "referred" is some constant plus a site effect. You have told SPSS that site 6 is the baseline, which means that site 6 is rolled into the intercept. The parameters for the other sites reflect how the response changes in comparison to what happens at site 6.
It would be nice to have an intercept term, representing the average response in some sense, and then parameters for each site ... however in such a model the 6 site effects would have to sum to the intercept term, making the design matrix singular. You would have one more parameter than your model strictly needs and the computer "will encounter a problem" when it attempts to estimate the parameters. To get around this technical difficulty, one chooses "reference levels" --- i.e., you fix one parameter to be 0, leaving you with 6 parameters to be estimated from the data. Other parametrizations are possible, but not, I think, in SPSS.
I don't know why SPSS is giving you numbers in the output. Did you code Site as a numeric with associated labels in the "Values" column? If so, you should get the label names on the output. And these names should orient you when you change the reference value to Site 1.
