elciclopeturnio
2011-Jan-03 13:32 UTC
[R] Greetings. I have a question with mixed beta regression model in nlme.
*Dear R-help: My name is Rodrigo and I have a question with nlme package in R to fit a mixed beta regression model. The details of the model are: Suppose that:* *j in {1, ..., J}* *(level 1)* *i in {1, ..., n_j}* *(level 2)* *y_{ij} ~ Beta(mu_{ij} * phi_{ij}; (1 - mu_{ij}) * phi_{ij}) y_{ij} = mu_{ij} + w_{ij} * *with* *logit(mu_{ij}) = Beta_{0i} + Beta_{1i} * x1_{ij} + b2 * x2_{ij} log(phi_{ij}) = Gamma_{0i} + Gamma_{1i} * z1_{ij} + c2 * z2_{ij} * *Beta_{0i} = b_0 + u_{0i} Beta_{1i} = b_1 + u_{1i} Gamma_{0i} = c_0 + v_{0i} Gamma_{1i} = c_1 + v_{1i} * *The vector* *(u_{0i}, u_{1i})'* *has normal distribution with mean* *(0, 0)'* *and covariance matrix* *sigma_{00} sigma_{01} sigma_{10} sigma_{11} * *The vector* *(v_{0i}, v_{1i})'* *has normal distribution with mean* *(0, 0)'* *and covariance matrix* *delta_{00} delta_{01} delta_{10} delta_{11} * *The* *w_{ij}s are independents. Each* *w_{ij}* *has mean* *0* *and variance * *W*. *The unknown parameters are:* * b_0, b_1, b_2, c_0, c_1, c_2, sigma_{00}, sigma_{10}, sigma_{11}, delta_{00}, delta_{10}, **and* *delta_{11}*. *The question is: How can I use nlme package in R to fit this model? If you want to know additional information, send me a mail, please. In advance, thanks a lot for your help. Kind regards, Rodrigo.* [[alternative HTML version deleted]]
Ben Bolker
2011-Jan-03 20:01 UTC
[R] Greetings. I have a question with mixed beta regression model in nlme.
elciclopeturnio <elciclopeturnio <at> gmail.com> writes:> > *Dear R-help: > > My name is Rodrigo and I have a question with nlme package > in R to fit a mixed beta regression model. The details of the model are: >[snip]> *The question is: > How can I use nlme package in R to fit this model? > If you want to know additional information, send me a mail, please.nlme only fits models with normally distributed residual error (and normally distributed random effects). lmer (in the lme4) package fits models with exponential-family individual effects (i.e. GLMMs), but that doesn't include beta I don't know offhand of any package in R proper that will fit this model, although you could try Jim Lindsey's 'repeated' package <http://popgen.unimaas.nl/~jlindsey/rcode.html> or one of the HGLM packages. There are R interfaces to WinBUGS and AD Model Builder (possibly among others). good luck Ben Bolker