[R] for help about logistic regression model

[Douglas Bates]

On 11/21/06, Aimin Yan <aiminy at iastate.edu> wrote:
> thanks, Here is data under this link with file name as p_5_angle.csv
> http://www.public.iastate.edu/~aiminy/data/
> p is protein names(5 proteins)
> aa are nested in p(up to 19 levels for each p, some p doesn't have 19 levels)
> index is position of aa.
> there are only one observation for each position of each aa within p.
>
> consider p as random effect,
> since aa is nested in p, so aa is also random effect.
>
> p and aa are qualitative predictors.
> x,y,z,sdx,sdy,sdz,delta,as,ms,cur are quantitative predictors.
> sc is binary responsible variable(>=90 and <90)
>
> we want to know the effect of p,aa,x,y,z,sdx,sdy,sdz,delta,as,ms,cur) on
> P(sc>=90).
>
> So I consider to use logistic regression model with p and aa as random effect.
>
> Firstly I try to use p,aa,x,y,z,sdx,sdy,sdz,delta,as,ms,cur as predictors,
> but it seems it has too many predictors.
> so I use p,aa,as,ms,cur as predictors, but it still doesn't work.

Here are the fits for two of your models using lmer.

> library(lme4)
Loading required package: Matrix
Loading required package: lattice
> p5 <- read.csv("http://www.public.iastate.edu/~aiminy/data/p_5_angle.csv")
> p5$Y <- p5$sc >= 90
> (mp5.NULL <- lmer(Y ~ 1|p/aa, p5, binomial, control = list(usePQL = FALSE)))
Generalized linear mixed model fit using Laplace
Formula: Y ~ 1 | p/aa
   Data: p5
 Family: binomial(logit link)
  AIC  BIC logLik deviance
 1390 1405   -692     1384
Random effects:
 Groups Name        Variance Std.Dev.
 aa:p   (Intercept) 0.447654 0.66907
 p      (Intercept) 0.015078 0.12279
number of obs: 1030, groups: aa:p, 92; p, 5

Estimated scale (compare to  1 )  0.9736361

Fixed effects:
            Estimate Std. Error z value Pr(>|z|)
(Intercept)  -0.1325     0.1151  -1.151     0.25
> (mp5.full <- lmer(Y ~ as*ms*cur + (1|p/aa), p5, binomial, control = list(usePQL = FALSE)))
Generalized linear mixed model fit using Laplace
Formula: Y ~ as * ms * cur + (1 | p/aa)
   Data: p5
 Family: binomial(logit link)
  AIC  BIC logLik deviance
 1278 1327 -628.8     1258
Random effects:
 Groups Name        Variance Std.Dev.
 aa:p   (Intercept) 0.085104 0.29173
 p      (Intercept) 0.026769 0.16361
number of obs: 1030, groups: aa:p, 92; p, 5

Estimated scale (compare to  1 )  0.9833564

Fixed effects:
              Estimate Std. Error z value Pr(>|z|)
(Intercept)  5.506e-01  1.714e-01   3.213  0.00131 **
as          -3.964e-02  2.322e-02  -1.707  0.08778 .
ms           1.879e-02  2.149e-02   0.874  0.38206
cur          3.413e-01  6.706e-01   0.509  0.61078
as:ms        1.091e-04  7.615e-05   1.432  0.15201
as:cur       8.315e-02  7.069e-02   1.176  0.23951
ms:cur      -4.880e-02  5.372e-02  -0.908  0.36366
as:ms:cur   -3.998e-04  6.602e-04  -0.606  0.54476
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
          (Intr) as     ms     cur    as:ms  as:cur ms:cur
as        -0.028
ms        -0.144 -0.960
cur       -0.391  0.070  0.019
as:ms      0.290 -0.655  0.443 -0.155
as:cur    -0.094  0.401 -0.288  0.211 -0.530
ms:cur     0.118  0.601 -0.672 -0.473 -0.185 -0.263
as:ms:cur  0.110 -0.354  0.232  0.001  0.614 -0.874  0.036
>