For my current research, I am trying to analyze a birdsong located on two islands. For ~20 years, it was considered as one species but two years ago they were split and considered as two. I wanted to determine whether the songs show variation throughout the years.

I used Raven Pro to analyze 1) the whole song and 2) the three parts of the song. For each song and its parts, I estimated 9 parameters: Duration 90% (s), Delta Time (s), Frequency 5% (Hz), Frequency 95% (Hz), Aggregate Entropy (bits), Bandwidth 90% (Hz), Center Frequency (Hz),s Delta Frequency (Hz), and Peak Frequency (Hz)

Aside from PCA, what other stat analysis I can do in R to fully maximize my dataset? Ideally, I would like to know the changes in the song in terms of the parameters taken as a whole throughout the years.

Would greatly appreciate any suggestions and comments. Thank you!

  • $\begingroup$ Can you provide more information on analysis done? Is analysis done on month by month, year by year, or on all 20 years lumped together? $\endgroup$
    – WMXZ
    Dec 14, 2022 at 7:59
  • $\begingroup$ This might be useless (it's not really my area), but it would not be too hard to compare the whole spectra of all individuals just checking for e.g. the R²-value for the goodness of fit. I'd suggest smoothing the spectra first. You "split"-criteria could be when R² drops below a certain value between the two sub-species (?) but check that the R² within the sub-species remains higher. $\endgroup$
    – Rasmus
    Dec 19, 2022 at 8:33
  • $\begingroup$ What about DFAs or some automatic classification tool? if there is statistical differences between the two islands, the computer will be able to separate them well, and then you can have a look at what is different... sorry for the late comment, I hope this is still up to date $\endgroup$
    – lframond
    Apr 3, 2023 at 7:28

1 Answer 1


As you wanted to determine the changes in time of bird song, the most basic analysis is a regression analysis. This gives you at least a feeling if there where changes and if these changes are significant. I guess, this can be done done with standard R.

You can carry out this regression analysis on each of the estimated parameters and on the PCA components.

Additional analysis may address island isolation questions, e.g. if the songs or part of it, are becoming simpler due to lack of competition. for this you may consider short term variances as new time series.

Other possible changes are variations not only in the note, but also in song characteristics (complexity in sequence of notes). This would require note classification as additional estimated parameter resulting in a temporal sequences of notes.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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