I recently worked on a consulting project where a client wanted to estimate gamma and weibull distributions based purely on histograms rather than raw-data. I have never worked with problems like that before, so if there are references (Books/articles) that people can point me to, that would be awesome.

  • 4
    $\begingroup$ Provided the histogram bins were predetermined (and not based on the data, as explained at stats.stackexchange.com/a/17148/919), maximum likelihood works well and is easy to apply. See examples at stats.stackexchange.com/a/34894/919 and stats.stackexchange.com/a/56100/919, for instance. These presume your histograms depict univariate datasets and not, for instance, data in some kind of regression problem. $\endgroup$ – whuber Jan 14 '20 at 15:39
  • 3
    $\begingroup$ Thanks! I'll take a look at that. We were looking at MLE-based methods to estimate and compare 2-parameter and 3-parameter weibull distributions using right-endpoints in histograms to measure wave heights, and we were struggling a bit. $\endgroup$ – Samir Rachid Zaim Jan 14 '20 at 15:47
  • 3
    $\begingroup$ Is the client saying the data aren't available? Or that the data are available but they believe the histogram is somehow superior to something that can be calculated from the full data/ $\endgroup$ – AdamO Jan 14 '20 at 16:51
  • 2
    $\begingroup$ Exactly - the data aren't available. They had some in-house data, but when they wanted to work with data from other agencies, the best they could work with was histograms from reports/publications. $\endgroup$ – Samir Rachid Zaim Jan 14 '20 at 18:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.