Timeline for Hypothesis testing on zero-inflated continuous data
Current License: CC BY-SA 4.0
14 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 17, 2023 at 4:24 | history | edited | User1865345 | CC BY-SA 4.0 |
added 43 characters in body
|
| Mar 25, 2013 at 19:47 | vote | accept | a11msp | ||
| Mar 25, 2013 at 19:37 | comment | added | user22507 |
@B_Miner Regarding the R-code for the combined approach (zero inflated gamma regression) you can use compound poisson models with the tweedie and the statmod package. Find the distribution parameter p (out$xi.max) with tweedie.profile and then fit the GLM with family=tweedie, e.g. summary(glm( y ~ x, family=tweedie(var.power=out$xi.max, link.power=0)))
|
|
| Mar 30, 2011 at 13:38 | comment | added | a11msp | I think I've found it: package gamlss provides model fitting based on several zero-inflated and adjusted continuous distributions. Thanks for your advice, B_Miner! | |
| Mar 30, 2011 at 7:06 | comment | added | a11msp | I work in biology, but yes - mine is a very similar problem. Will try to find an r implementation of the SAS code you mentioned. I'm also wondering if the r package cozigam is doing a similar thing. Will investigate. | |
| Mar 29, 2011 at 21:31 | comment | added | B_Miner | For example, I work in direct marketing and so want to use a model to test the differences between two direct mail versions. I am interested in the response rate difference (P(Y=0|X)) and the average revenue of those who order (P(Y|X; Y>0)) yes, but at the end of the day I am concerned about the expected revenue value of those that received version A versus B and can not use a simple ANOVA because of the (large amount = those that did not order) zeroes. Again, don't know your domain but hope this helps. | |
| Mar 29, 2011 at 21:31 | comment | added | B_Miner | @msp: If you can fit the model shown in the link I supplied (SAS) in either SAS or R (I am sure there is an NLMIXED analog in R, especially since this model uses only fixed effects, it's just that SAS does not have any other procedure to write your own log-likelihood) you can accommodate a hypothesis test using P(Y|X) instead of separate hypothesis using the two models in the two-stage approach. | |
| Mar 29, 2011 at 19:32 | comment | added | B_Miner | Yes, the two stage approach (Gelman method of two seperate models) assumes two populations, those at zero and those > 0. Regarding the hypothesis tests, can you frame them in terms of the predicted values for varying levels of the inputs and construct empirical confidence intervals related to the simulations for each? For hypothesis tests for the coefficient != 0, you need to test this seperately for both models. | |
| Mar 29, 2011 at 19:28 | comment | added | a11msp | ...so would it be appropriate to simply say that if the impact of some factor level is significant (and significantly different from that of some other factor level) in either of the two models in Gelman's method then it's significant overall? | |
| Mar 29, 2011 at 19:25 | comment | added | B_Miner | The two stage approach (Gelman method of two seperate models) assumes two populations, those at zero and those above. | |
| Mar 29, 2011 at 19:02 | comment | added | a11msp | I've now replicated the two-stage approach proposed in the Gelman book. If subset_factor (with 25 levels) serves as subset label, the first step is fit1=glm(response~subset_factor, family=binomial); and the second step is fit2=lm(response~subset_factor, subset=response>0). I can then run simulations as they describe to obtain the distribution of fitted response values for each factor level. However, I'm still unsure how to translate this to what I need, which is (a) the probability that coefficients aren't zero and (b) significance of difference between coefficients at different factor levels. | |
| Mar 29, 2011 at 18:48 | vote | accept | a11msp | ||
| Mar 25, 2013 at 19:47 | |||||
| Mar 29, 2011 at 17:42 | comment | added | a11msp | @B_Miner, thanks very much for your answer, sorry don't have enough rating to vote you up... I'll have a look at the links! My only worry about conditional models is that they postulate that zeros cannot belong to the second (continuous) component, am I right? Does my setting not feel a bit more like a mixture model? What do you think? | |
| Mar 29, 2011 at 17:17 | history | answered | B_Miner | CC BY-SA 2.5 |