[R-sig-ME] gee, geese and glmer

Yang, Qiong qyang at bu.edu
Mon Mar 31 16:09:47 CEST 2014


Hi Ben,
I did some exploration of the data. There is a substantial number of cases who are singleton (sorry I thought all cases are singleton in last email).
In overall sample, the crude odds ratio (OR) ignoring family structure is negative consistent with old lme, GEE. New lme4 returns a positive OR.
	All			
	male(1)	female(2)		
0	2554	3021		OR=
1	310	290		0.790872495
				
If we only include singleton cases -versus family controls, the crude OR is negative.		
	male(1)	female(2)		
0	2554	3021		OR=
1	249	182		0.617934166
				
If we only include family cases-versus family controls, the crude OR is positive.		
	male(1)	female(2)	
0	2554	3021		OR=
1	61	108		1.496801081

I wonder if singleton cases are dealt differently between new and old lme4 that causes different signs of sex in the results?

Thanks
Qiong 
----Original Message-----
From: Ming-Huei Chen [mailto:mhchen at bu.edu] 
Sent: Tuesday, March 18, 2014 10:35 PM
To: 'Ben Bolker'; 'Vincent Dorie'
Cc: Yang, Qiong; 'r-sig-mixed-models'; 'lme4-authors at lists.r-forge.r-project.org'
Subject: RE: [R-sig-ME] gee, geese and glmer

I just installed latest lme4. Here is the result... 

>
blme::bglmer(case~sex+PC1+PC2+PC3+PC4+(1|famid),family=binomial,data=test.da
t,fixef.prior=t)
Cov prior  : famid ~ wishart(df = 3.5, scale = Inf, posterior.scale = cov, common.scale = TRUE) Fixef prior: t(df = 5, scale = c(100, 0, 0, 0, ...), common.scale = FALSE) Prior dev  : 16.1703

Generalized linear mixed model fit by maximum likelihood (Laplace
  Approximation) [bglmerMod]
 Family: binomial ( logit )
Formula: case ~ sex + PC1 + PC2 + PC3 + PC4 + (1 | famid)
   Data: test.dat
      AIC       BIC    logLik  deviance  df.resid
 3005.889  3052.986 -1495.944  2991.889      6168
Random effects:
 Groups Name        Std.Dev.
 famid  (Intercept) 19.89
Number of obs: 6175, groups: famid, 1592 Fixed Effects:
(Intercept)          sex          PC1          PC2          PC3          PC4
   -10.2654       0.3371      -0.3533      -0.1078      -0.2225       0.1102

-----Original Message-----
From: Ben Bolker [mailto:bbolker at gmail.com]
Sent: Tuesday, March 18, 2014 4:31 PM
To: Ming-Huei Chen; 'Vincent Dorie'
Cc: 'Yang, Qiong'; 'r-sig-mixed-models'; 'lme4-authors at lists.r-forge.r-project.org'
Subject: Re: [R-sig-ME] gee, geese and glmer

On 14-03-18 02:51 PM, Ming-Huei Chen wrote:
> Is lme4_1.1-4 out-of-date? This is the version I used and encountered 
> the error.
> 
> Best,
> 
> Ming-Huei

  Yes, 1.1-5 was released on CRAN recently (and blme did some updating to match).

  Ben Bolker

> 
> -----Original Message-----
> From: Vincent Dorie [mailto:vjd4 at nyu.edu]
> Sent: Tuesday, March 18, 2014 11:48 AM
> To: Ben Bolker
> Cc: Yang, Qiong; Ming-Huei Chen; r-sig-mixed-models; 
> lme4-authors at lists.r-forge.r-project.org
> Subject: Re: [R-sig-ME] gee, geese and glmer
> 
>>>
>>> blme::bglmer(case~sex+PC1+PC2+PC3+PC4+(1|famid),family=binomial,data
>>> =
>>> test.da
>>> t,fixef.prior=t)
>>> Error in get("checkConv", lme4Env)(attr(opt, "derivs"), opt$par, 
>>> ctrl = control$checkConv,  :
>>>  unused argument (ctrl = control$checkConv)
> 
> I believe this can only happen if the installed version of lme4 is 
> out-of-date. If there was a point where checkConv was added but didn't 
> yet have the ctrl formal, you would see that error.
> 
> Vince
> 



More information about the R-sig-mixed-models mailing list