I want to do a regression using 2 or 3 predictor variables, but the data for each variable will be available for several years; most likely 2008-2014. Is this called a time-series regression? I have learned multiple regression as a psych undergraduate, but with multiple years I'm not quite sure where to begin. If I'm on the right track with time-series, can anyone direct me to an SPSS based tutorial?
Thanks in anticipation!
This sounds like a mixed model; you could have time and the predictor (and their interaction) as fixed effects, and Subject (or whatever was tested at each year) as a random effect (with maximal random slopes if possible). In R this can be done with the lmer() function from the {lme4} package; I know it's also possible in SPSS but I haven't used that in a long time so I don't remember the precise procedure there.
Andrew -
You might be looking at "panel (i.e., longitudinal) data."
Regardless, you need to decide the role of time, if any, considering subject matter application and theory.
You also need to know if you are following a time series on a single entity, or various entities, or are you really looking at a cross-sectional data set at some given time.
To distinguish between a time series, and a cross-sectional survey, you might consider the following:
https://www.researchgate.net/publication/275365705_When_Prediction_is_Not_Time_Series_Forecasting
A panel survey contains aspects of each.
In addition, you might want to look at what might be a related question and answers:
https://www.researchgate.net/post/How_do_I_set_time_variables_in_estimating_crop_yields
Best wishes - Jim
PS - Looking at your question again, I think you do want to look at panel survey literature. Panel surveys are not a particularly 'big' area for me, however, and I can't comment on SPSS.
If you are estimating for finite populations, perhaps totals, however, then you might just be considering each year separately, and then comparing them. That is what is often done in agencies responsible for official statistics, where there is often no real "time series" being considered, in a mathematical sense, just periodic results from finite populations censused at times and sampled more frequently.
Just be sure that the methodology you use is consistent with the data you collect, how you collect it, and your goals. - Best wishes.
Research When Prediction is Not Time Series Forecasting
This sounds like a mixed model; you could have time and the predictor (and their interaction) as fixed effects, and Subject (or whatever was tested at each year) as a random effect (with maximal random slopes if possible). In R this can be done with the lmer() function from the {lme4} package; I know it's also possible in SPSS but I haven't used that in a long time so I don't remember the precise procedure there.
Thank you both for your helpful answers - looks like I have some reading to do!
Andy,
I don't think SPSS does Panel data. A windows based (and easy) user-friendly package to do this sort of thing is Eviews.
Best
Phil
SPSS can run linear mixed models (aka multilevel model)and time series analyses (though I no longer use it myself). If you have experience with multiple regression I'd think about using a linear mixed model with covariance structure to account for autocorrelation e.g., AR(1) etc.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis