I want to use chronotype as a covariate in a regression model. My chronotype variable are time points; .i.e. the mid-sleep point corrected for sleep debt. STATA doesn't like this and when it is run in a regression model, the data is seemingly converted to exponentials. I have found papers before which have used chronotype as a covariate in their models but I was wondering if the variable was actually converted to a numeric format or changed to be able to be used in regression analyses. Or, is there a way on STATA for indicating that it is a time (not time-series).
Any help regarding how chronotype looks as an actual variable when being used in regression analyses would be extremely helpful. Thanks
Some software cannot handle the clock times (e.g., alongside Stata, also Mplus, while SPSS can handle the clock times in regression analyses). However, in STATA or Mplus, this can be easily solved. For your analysis in STATA, I would suggest to convert the clock times to a numerical format, for example, convert midpoint of sleep (MSF) to seconds (e.g., for 1 day you get values between 0 and 60 * 60 * 24 = 86400 [a person with a MSF of 1:00am has a value of 3600, a person with a MSF at 2:30am has a value of 9000]; if some of your respondents have a MSF before midnight, cover two days = 48 hours, so you get values betwen 0 and 172800 [a person with an MSF at 1:00am has a value of 86400 + 3600 = 90000, a person with a MSF at 2:30am has a value of 86400 + 9000 = 95400]).
The conversion should be easy, at least in SPSS you just change the format from date to numeric (I don't know how this is handled in STATA).
I hope this helps.
We need a bit more information here; what are your other variables, how is time points exactly defined in your data set?
Best, CCC
Thank you very much Christian. This is amazingly helpful.
Claus-Christian - Currently I have other time parameters such as TST, WASO and other sleep parameters. I also have ordinal variables that are derived from other sleep questions. In the future I will have further ordinal and binary data that are derived from psychiatric assessments. The intention is for the MSF/MSFsc to be a predictor in regressions of psychiatric behaviour but until I converted the scores to seconds, STATA was unable to tolerate this.
Ah I never experienced this, but I was using SPSS / R. The suggestion Christian made is probably the best, you could also try including MSFsc as a continuous variable. For instance, when you have a MSFsc on 3:40, it becomes 3,66. 2:30 becomes 2,5.
Good luck!
Hi Stefan,
Thanks for your suggestion. I have done both of these options to have a look at which is more manageable as a variable before you said, so great minds think a like! Thank you for your comment.
- Badges
- Science topic
- Similar topics
- Biological Science
- Physiology
- Physiological Processes
- Stress