I'm very new to statistics and I've been asked to create a time series model in SPSS. I've taken a list of articles from one journal, each categorized by their main topic (topics have been numerically coded). The journal has been reanalyzed every 5 years.

enter image description here

I'm supposed to determine how these topics would vary with time, but I'm not sure how to input this into the Forecasting dialogue, and anyway the 'Topics' variable doesn't even denote real numerical value, which seems to be the case for every forecasting tutorial I've read. Is time-series even the best option for this kind of data?

  • 1
    $\begingroup$ This question isn't about how to use SPSS; it's about how to model a time series of multinomial observations. This should stay open, IMO. $\endgroup$ – gung - Reinstate Monica Sep 8 '16 at 13:02

Time series forecasting generally refers to numerical rather than categorical variables. They are also generally used with longer timeframes - based on what I can see, your dataset extends from 1970 to 2015 at the most.

You could do some simple visualizations for a starter, to see if there are any obvious patterns. For example, you could consider creating a line graph for each topic. Count the number of times each topic appears in each 5-year interval (4 times for topic 3 in 1990, based on what I can see) and then look at how that changes over time.

You could also transform your categorical variables into numerical ones by turning them into percentages, such as the percentage of journals that were about a specific topic. That would allow you to use more statistical methods (such as ARIMA / ARCH).

| cite | improve this answer | |
  • $\begingroup$ Thanks for the suggestions! As well, do you think it'd be a good idea to run a multinominal logistic regression on this data set? I'm considering adding in other variables such as author name and author country. $\endgroup$ – Lora Jun 20 '16 at 18:36
  • $\begingroup$ @Lora Unfortunately, I'm not particularly familiar with multinomial logistic regression, so I can't say. I'm actually not sure if logistic regression would be the right way to go at all with this dataset - it's entirely possible that linear regression will be enough to adequately model this, especially if you use a time series model. $\endgroup$ – Kontorus Jun 20 '16 at 18:51
  • $\begingroup$ I do not know if the term "forecasting" is used, but note that Hidden Markov Models are commonly used for prediction in sequences of categorical variables (including time series). $\endgroup$ – GeoMatt22 Sep 8 '16 at 13:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.