I have a dataset that includes observations at different time points. There are multiple observations at the same time points and the time points are not evenly spaced.

Now, I would like to test whether the data are random or autocorrelated in time. Can I call such a dataset as a time series? If so, what kind test methods I can try to use for such an irregularly spaced time series?


While it's apparently some form of time series, if there are multiple observations at a single point, it might be a little unusual to call it a time series.

One difficulty I see is how do I relate (say) 2 observations at time $\tau$ to 3 observations at time $\tau-s$? What's the model? Does the expectation at time $\tau$ relate to the average at time $\tau-s$ or what?

With single-observation-at-a-time, there's certainly ways to write continuous-time autocorrelated models.

A common example is the Ornstein-Uhlenbeck process. See also the Vasicek model, a particular example of its use.

It's possible to write an unevenly-sampled observational model based on the O-U process as something of the form (if $s,t$ are consecutive observation times):

$\hspace{1cm}(y_t-\mu) = \phi^{t-s} (y_s-\mu) + n_{t-s}\,,\quad$ where $\:n_u\sim N(0,\frac{\sigma^2}{2\phi} (1-e^{-2\phi u}))$

(I think!), which when $t-s=1$ (i.e. the usual regular-time interval situation) corresponds to an AR(1).


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.