2
$\begingroup$

I have time-series data. It includes a number of patients with series conditions due to car crashes. After I remove the trend, I found that some number of patients becomes negatives which is impossible. What is the problem? What is my mistake? I remove the trend because I do not want to model it as a covariate. Why? because, I read that if I do not want to model the trend as a covariate, then I can remove it. What I should do to remove the negative values?


Should I remove the trend of this data? If so, how can I avoid negative values when removing the trend?

$\endgroup$
  • $\begingroup$ After you model it, you can add the trend back. $\endgroup$ – kevin012 Jul 21 at 10:37
  • $\begingroup$ @kevin012 Thanks a lot for your comment. Do you mean that I can model by data and then return back the trend for it to predict the future values (regression)? $\endgroup$ – Maryam Jul 21 at 11:31
  • $\begingroup$ Yes, because you have a linear trend, it would not be difficult to add the trend back. If there is still negative value and if it's not serious, you may force a lower bound for your forecast. $\endgroup$ – kevin012 Jul 21 at 11:47
  • $\begingroup$ @kevin012 can I keep the trend. $\endgroup$ – Maryam Jul 21 at 15:35
  • $\begingroup$ What do you mean by keeping the trend? $\endgroup$ – kevin012 Jul 21 at 22:09
2
$\begingroup$

Certain time series methods, like ESM deal fine with trends. Others, like ARIMA don't. So in some cases the method requires you to remove it and sometimes not. I think you should focus more on why you would want to remove it or not assuming your method allows you to have it. Sometimes you want to separate out seasonality, trends etc so you can see them.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ My method do not require to remove the trend. However, some researcher said removing the trend improve the model fit. $\endgroup$ – Maryam Jul 23 at 10:18
  • $\begingroup$ You would have to know why they believe this and if their assumptions are likely true. Obviously if you believe the actual data has a trend in it choosing the right model is important (as is true in all regression). In ESM if there is a trend you should chose an ESM model that assumes trend. $\endgroup$ – user54285 Jul 23 at 17:40

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.