3

I am estimating a zero/one inflated beta regression model with gamlss (family BEINF). My dependent variable is [0,1] with a lot of 0s, quite some 1s, and some values in between. This means I assume that my model consists of three submodels:

enter image description here

Reference formula: Rigby & Stasinopoulos (2010), A Flexible regression approach using GAMLSS in R, section 10.8.2, p 215.

With regard to link functions used, this source gives a nice table (see BEINF)

enter image description here

In R, the coefficients of these submodels are addressed by Nu (zero-probability), Mu and Tau (one-probability)?

gam<-gamlss(y_proportion ~ x1+x2, nu.formula=~x1+x2, tau.formula=~x1+x2, family= BEINF,data=Alldata)
### find raw probabilities
prob.of.0.raw <- predict(gam, what="nu", type="response")
prob.of.1.raw <- predict(gam, what = "tau", type = "response")
prob.of.mu <- predict(gam, what="mu", type="response")[1]

My question is related to Mu. What exactly is this?

  • Some people tell me this is the "mean". But do they mean the "mean" of the entire model (all three of them together) and is it therefore comparable with mean(y).
  • Or is it only the mean of the middle model (and is it therefore not taking into account 0 and 1)? I am assuming it is the last answer but I am confused. Thanks for your help!
Scortchi - Reinstate Monica
  • 27,560
  • 8
  • 81
  • 248
user33125
  • 163
  • 2
  • 13
  • Might be easier to answer if you gave clearer references. The first formula (from where?) uses a completely different set of parameters on the left side to on the right side! What does the`gamlss` documentation say? – Scortchi - Reinstate Monica Dec 29 '16 at 20:37
  • Oh snap! I thought I gave the references. I will add them, thanks! The gamlss is only explaining Tau and Nu (as far as I know because it is a big manual). – user33125 Dec 29 '16 at 20:39
  • I also added a table for the parameters on the left side. I have always understood these as an indication of the distribution used for the different submodels. – user33125 Dec 29 '16 at 20:48
  • 1
    The definitions of the parameters $\mu$ &c. are given immediately below Equation 10.47. I'd suggest you include them in your question, as they;re essential to answering it. – Scortchi - Reinstate Monica Dec 29 '16 at 20:48
  • 1
    The table's of the link functions used. – Scortchi - Reinstate Monica Dec 29 '16 at 20:51
  • 1
    Thanks for the suggestions/clarifications. I edited the question. Hope it is clear now! – user33125 Dec 29 '16 at 20:56
  • 1
    Thank you. It's clear, I believe. Briefly, $\mu$ is the mean of the [beta distribution](https://en.wikipedia.org/wiki/Beta_distribution), the middle model. The mean of the entire model is given by $\operatorname{E}{y}=\frac{\tau + \mu}{1+\nu+\tau}$. An answer will go into more detail. – Scortchi - Reinstate Monica Dec 29 '16 at 21:16
  • I was hoping that would be the correct answer! Now I can continue with more confident! Thank you very much! – user33125 Dec 29 '16 at 21:43
  • I would like to reopen this post because I think it is related with the named nu parameter. I am trying to understand what is nu in the family = BEZI. The named references here (Rigby & Stasinopoulos (2010), A Flexible regression approach using GAMLSS in R, section 10.8.2, p 215.) only explains this for BEINF, BEINF0 and BEINF1, but what about BEZI and BEOI? Is it the same? Which linear predictor is nu exactly? – Elvira Nightingale May 31 '21 at 16:57
  • Welcome to the site. Please do not use the "Your Answer" field to ask new or follow-up questions. Instead, click the blue [ASK QUESTION](https://stats.stackexchange.com/questions/ask) link at the top of the page & ask there. If you want to address the author of an answer, you will need 50 reputation first to leave comments. Since you're new here, you may want to take our [tour], which has information for new users. – gung - Reinstate Monica May 31 '21 at 17:07

0 Answers0