For my master thesis I conducted an online experiment where participants had to conduct a shopping task where they were provided with a local and a non-local product three times in a row. So for three times, they had to choose between either the local or the non-local product. The result of the shopping task is my dependent variable "Green Shopping Behavior". Every time someone chose the local product, they got a "1" and if they chose the non-local product they got a "0". In the end i added everything up, so for each observation the dependent variable can take the values 1,2 and 3. Originally i thought that my DV is scaled metrically, since the distance between the values is the same and can be quantified. Since I wanted to run a linear regression and have some problems with that I am now asking myself if my variable is even metrically scaled. Could you help me with that? Best regards Carina
1 Answer
These are basically counts, so measurement level is not a problem, these can be interpreted as metrically, interval, or even ratio scaled (some would even call counts an absolute scale).
Note though that just because your variable is scaled metrically, this doesn't mean that a linear regression will work fine with it. Note that generally statistical model assumptions are in terms of distributions and not in terms of measurement scales. Distribution and measurement scale level do not imply each other. I'm not sure whether you want to use your variable as predictor or response, but a response with just 4 possible values (it can also take 0, can't it?) is usually not very well suited for linear regression (note that regression responses are in theory unbounded!). I don't see a problem with such a variable as a predictor.
For response variables of this kind consider Binomial regression.
-
$\begingroup$ Hey Christian, thank you very much for your answer - it was very helpful. You are right that i have count data and that i have 4 possible values. That a linear regression isn't a good model fit is what i expected/concluded from testing the assumptions and looking at plots. It is was what i proposed in my expose along with my variables thus i was a bit confused. I had a look into binomial regression and think you led me on the right track but i still have some question - it would be highly appreciated if you would take the time to clarify them for me. $\endgroup$– carinaCommented Sep 20, 2022 at 9:13
-
$\begingroup$ The variable I described is my dependent variable “Green Shopping Behavior” which can take on the values= 0, 1, 2, 3. I have 2 metric/continuous independent variables i want to use to predict if participants chose 0/1/2/3 local products, as well as control variables. I am using STATA for my project. Looking into binomial regression, i ran into binomial logistic regression and for me it looks the same (despite being called differently). I further stumbled across the term “generalised linear models” which i understood is a general term under which binomial (logistic) regression falls. $\endgroup$– carinaCommented Sep 20, 2022 at 9:13
-
$\begingroup$ Looking at options in STATA, i can run a logistic regression using the command “logistic” or i can run generalised linear models using different versions of the “glm” command. Could you clarify for me what would be the best strategy? Thank you very much! $\endgroup$– carinaCommented Sep 20, 2022 at 9:14
-
$\begingroup$ @carina Sorry, I don't do STATA. $\endgroup$ Commented Sep 20, 2022 at 11:50
-
$\begingroup$ Hey Christian, could you point me towards the "right" method either way? i am still confused about which regression would be right - i am currently looking into ordered logistic regression but am unsure on how to proceed. would be highly appreciated! $\endgroup$– carinaCommented Sep 20, 2022 at 12:13