Does it make sense to enter date (min:second) as dependent variable in multiple linear regression model? If yes, how can interpret its regressions coefficient?
I would advise against doing this but rather turn the whole thing into elapsed time since start of the process in say seconds and then if necessary (not to get huge absolute values that can be numerically problematic) divide by say 1000. I like to get the response in the range of 0 to 100 by deflating in this way. You could of course divide by 60 to get minutes and decimal parts of minutes.
I would not do any modeling until I had plotted the data.
I would advise against doing this but rather turn the whole thing into elapsed time since start of the process in say seconds and then if necessary (not to get huge absolute values that can be numerically problematic) divide by say 1000. I like to get the response in the range of 0 to 100 by deflating in this way. You could of course divide by 60 to get minutes and decimal parts of minutes.
I would not do any modeling until I had plotted the data.
Yes, as Kelvyn said:
use "time since start (of the experiment)" in seconds, or in minutes, or in hours...
If you decide to use for instance "time since start in seconds", then the meaning of the coefficient will be: "expected change in the response per second".* If you use the conventional reference coding, the intercept of your model gives the "expected response value at the start of the experiment".
--
*This is only that simple when the model is that simple, i.e when you model the response as linearily depending on the time. This might not fit well to your data - you have to check this: plot your data, see if it is a linear relationship. Sometimes relationships might not be linear. In such cases transformations (of the response and/or the time values) can linearize the relationship. The interpretation of the coefficient is then obviousely with respect to the transformed variables. If you use polynomials to fit non-linear relationships, the interpretation of individual coefficients is difficult and not very intuitive.
I see you have SPSS listed as one of the topics, so assume that is the software you are using. You may find the DATEDIFF function very useful for computing elapsed time from some starting point. See the link below for more info including examples.
HTH.
http://www.ibm.com/support/knowledgecenter/SSLVMB_20.0.0/com.ibm.spss.statistics.help/syn_date_and_time_date_differences.htm
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS