I re-read this article after a long time (this time paying attention to the fine print) and noticed that they use indel costs that depend on sequence length in their OMA. They argue that it is done to minimise the influence of sequence length in their dataset, because much of these differences are due to censoring rather than to true difference in sequence length. Their setup is so that sequences of different lengths are compared with an indel cost of roughtly 1/4 of the indel cost used for sequences of equal length. They use a self-written SAS program for that.

I reckon that the paper is a bit old and that much has come out since then. Is that issue still (or was it ever) relevant? If so, how can I implement such algorithm using R or Stata?


1 Answer 1


As we argue in our review of sequence dissimilarity measures, I would not recommend this distance measure, because it breaks the triangle inequality. Depending on the distance analysis procedure (clustering) you plan to use, this may be problematic.

  • $\begingroup$ Thank you for your response. Coincidentally I was reading your paper right now. I also use TraMineR, so could you help me with choosing an algorithm that deals with the problem Stovel and Bolan tried to solve without violating triangle inequality? $\endgroup$
    – Kenji
    Commented Dec 11, 2015 at 13:21

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.