Legend

849 reputation

720

bio	website	enigmaportal.com
	location	United States
	age	27
visits	member for	2 years, 5 months
	seen	May 8 at 4:38
stats	profile views	139

Just a plain old simple guy :)

	849 reputation	bio	website	enigmaportal.com	visits	member for	2 years, 5 months
720 badges		location	United States		seen	May 8 at 4:38

summary answers questions tags badges favorites bounties reputation activity

285 Actions

suggestions reviews revisions comments badges posts accepts all

1d	awarded	Taxonomist
May 7	revised	Fitting mixture distributions and computing goodness-of-fit? added 394 characters in body
May 7	comment	Fitting mixture distributions and computing goodness-of-fit? @rbatt: +1 Yes. I am interested in learning more about this topic from a practical perspective and thought this really is a very different topic so I'm posting it as a new question. I am doing some reading but due to my lack of experience, I don't want to use these techniques incorrectly. :) I'll add a cross-reference to the previous question. Thanks for pointing it out.
May 7	asked	Fitting mixture distributions and computing goodness-of-fit?
May 7	comment	What distribution does my data follow? My bad! You are right. They both indeed show the four peaks. Thank you once again for your time.
May 7	comment	What distribution does my data follow? @Glen_b: Sorry. Maybe I am going off-track here. I saw something here: stats.stackexchange.com/questions/28873/… Also, what apparently seemed like a peak using the Box Cox transformation no longer appears in logspline. Or am I to think that these transformations may find mutually exclusive peaks? I plotted the logspline but cannot figure out how to obtain the x and y values from the object to examine this in detail.
May 7	comment	What distribution does my data follow? @Glen_b: Great! The `logspline` seems very interesting and pin points these times in a much clearer way than the fourth root transformation. I'll examine these in more detail. I was using `rebmix` like this:`REBMIX(Dataset = list(duration=t), Preprocessing = c("histogram", "Parzen window"), cmax=4, Criterion = c("AIC", "BIC"), Variables="continuous", pdf="lognormal", K=7:20, b=0)` I am assuming I need now write a function to compute the KS value to understand the fit of this mixture?
May 6	comment	What distribution does my data follow? @rbatt: Definitely! In the mean time, I hope you won't mind me bugging you a bit more: how does one do this today? That is, attempting to fit multiple distributions? Is there a way to formally split the dataset and say that, this set belongs to the first distribution and this set belongs to the second? Can you show me an example of how this analysis is done?
May 6	comment	What distribution does my data follow? @rbatt: +1 Thank you. Is there an R library that I can leverage to do this or should this analysis be fairly hand-driven?
May 6	comment	What distribution does my data follow? @Glen_b: Great! Thank you very much. I will read the references you provided.
May 6	comment	What distribution does my data follow? +1 Thank you for the reference and I apologize for the usage. I did not mean to disrespect anyone. The library I was using `fitdistrplus` called it a Cullen-Frey graph and I continued using it.
May 6	comment	What distribution does my data follow? Forgot to accept this as the answer. I learnt a lot from your post. Thank you once again. My previous question still holds though: I'd really appreciate if you can suggest some book/reference that I can use to learn this "detective" work :)
May 6	accepted	What distribution does my data follow?
May 6	comment	What distribution does my data follow? +1 This is a gold mine of information of me. I am trying to digest everything you have written and so far this has taught me how to actually approach this type of problems. What is the point of the stronger transformation? May I ask how you came up with that? Is that with experience or is there a more formal way of choosing such a non-conventional transformation? Please pardon my ignorance if this is common wisdom in the stats community. But I would be thankful if you could point me to a good reference to learn this kind of "detective" work which feels awesome to me.
May 6	comment	What distribution does my data follow? @Glen_b: Great! Thank you for your time.
May 6	comment	What distribution does my data follow? Also, updated my question with the data I'm using at the end in case it helps.
May 6	comment	What distribution does my data follow? @Glen_b: Thank you. I just included a qqplot for lognormal as well but even this does not seem to be a good fit. Is there anything else you recommend that I try out? I included my data in the question.
May 6	comment	What distribution does my data follow? And one more for the lognormal distribution. Do you recommend any pre-processing that I should do with the data? Or is there a better way to estimate the best-fit? I'm still wondering how I can utilize the Cullen/Frey graph in my context.
May 6	revised	What distribution does my data follow? added 133 characters in body
May 6	comment	What distribution does my data follow? Just updated my question with a `qqPlot` from the `car` package.