I have apartment data on rents from 2000 to 2016 where the unit of observation is an individual property. The sampling methodology included more properties as time went on thereby expanding the original sample over time.

My question is whether there is any statistical implication to using data where n increases as t increases. Does any bias occur as a result? Is it best to limit the sample to repeat properties only?

  • $\begingroup$ For clarification: Once an apartment enters the data base does it stay there for all remaining years? $\endgroup$ – Michael Chernick Jul 30 '18 at 22:51
  • $\begingroup$ Can you give us the number of apartments that are included in each year? It would be helpful to know to what degree the number of apartments grows. Technically this sounds more like a longitudinal study. A simple time series would have one observation at each time point and would show how the rent varied over time for that apartment, $\endgroup$ – Michael Chernick Jul 30 '18 at 22:57
  • 1
    $\begingroup$ I am going to edit your question. If I get something wrong feel free to make the changes that you want. $\endgroup$ – Michael Chernick Jul 30 '18 at 22:58
  • 2
    $\begingroup$ This is question is not clear at all. We can't tell you the effect of sample size until you tell us what you are trying to model. What type of analysis/modelling are you trying to pull off? $\endgroup$ – Skander H. Jul 30 '18 at 23:09

According to Wikipedia...

Ideally, unevenly spaced time series are analyzed in their unaltered form. However, most of the basic theory for time series analysis was developed at a time when limitations in computing resources favored an analysis of equally spaced data, since in this case efficient linear algebra routines can be used and many problems have an explicit solution. As a result, fewer methods currently exist specifically for analyzing unevenly spaced time series data.

A common approach to analyzing unevenly spaced time series is to transform the data into equally spaced observations using some form of interpolation - most often linear - and then to apply existing methods for equally spaced data. However, transforming data in such a way can introduce a number of significant and hard to quantify biases, especially if the spacing of observations is highly irregular.

However, don't fret, there are plenty of ways to analyze irregularly-spaced time series data. Some of them are definitely better than others, depending on the outcome you would like to observe (theoretical relationships between variables, predictive ability, etc.). This PDF gives a decent mathematical basis for the problem you are struggling with.

If you like using MATLAB, you may find this intro to resampling helpful. The example explains how to resample nonuniformly sampled signals to a new uniform rate.

You also may find some of the R packages at the end of this whitepaper helpful, particularly the packages mentioned in section 7.


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.