search

bwallace / OpenMeta-analyst-

intuitive software for advanced meta-analysis
39 stars 19 forks source link

Allow user to set the number of integration points in bivariate #243

Closed bwallace closed 10 years ago

bwallace commented 10 years ago

In bivariate.r, line 52 there is a call out to the bivariate package including this parameter: nAGQ = 3. This is the number of integration points. This should be a parameter to the method (so ultimately added to the wrapper method). The default should still be 3, I think.

gdietz commented 10 years ago

I disagree with this suggested enhancement. The error message we get when trying to use nAGQ = 3 (i.e. nAGQ != 1) is:

Error in updateGlmerDevfun(devfun, glmod$reTrms, nAGQ = nAGQ) : nAGQ > 1 is only available for models with a single, scalar random-effects term

Tom claims he doesn’t known what this means. Thus in lieu of a tom understanding, its foolhardy to try anything else.

On Nov 14, 2013, at 7:56 AM, byron wallace notifications@github.com wrote:

In bivariate.r, line 52 there is a call out to the bivariate package including this parameter: nAGQ = 3. This is the number of integration points. This should be a parameter to the method (so ultimately added to the wrapper method). The default should still be 3, I think.

— Reply to this email directly or view it on GitHub.

bwallace commented 10 years ago

ok -- just make it 1 then?

On Fri, Nov 15, 2013 at 11:54 AM, George Dietz notifications@github.comwrote:

I disagree with this suggested enhancement. The error message we get when trying to use nAGQ = 3 (i.e. nAGQ != 1) is:

Error in updateGlmerDevfun(devfun, glmod$reTrms, nAGQ = nAGQ) : nAGQ > 1 is only available for models with a single, scalar random-effects term

Tom claims he doesn’t known what this means. Thus in lieu of a tom understanding, its foolhardy to try anything else.

On Nov 14, 2013, at 7:56 AM, byron wallace notifications@github.com wrote:

In bivariate.r, line 52 there is a call out to the bivariate package including this parameter: nAGQ = 3. This is the number of integration points. This should be a parameter to the method (so ultimately added to the wrapper method). The default should still be 3, I think.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHubhttps://github.com/bwallace/OpenMeta-analyst-/issues/243#issuecomment-28584503 .

byron

gdietz commented 10 years ago

already done, care to remove this issue then?

On Nov 15, 2013, at 12:00 PM, byron wallace notifications@github.com wrote:

ok -- just make it 1 then?

On Fri, Nov 15, 2013 at 11:54 AM, George Dietz notifications@github.comwrote:

I disagree with this suggested enhancement. The error message we get when trying to use nAGQ = 3 (i.e. nAGQ != 1) is:

Error in updateGlmerDevfun(devfun, glmod$reTrms, nAGQ = nAGQ) : nAGQ > 1 is only available for models with a single, scalar random-effects term

Tom claims he doesn’t known what this means. Thus in lieu of a tom understanding, its foolhardy to try anything else.

On Nov 14, 2013, at 7:56 AM, byron wallace notifications@github.com wrote:

In bivariate.r, line 52 there is a call out to the bivariate package including this parameter: nAGQ = 3. This is the number of integration points. This should be a parameter to the method (so ultimately added to the wrapper method). The default should still be 3, I think.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHubhttps://github.com/bwallace/OpenMeta-analyst-/issues/243#issuecomment-28584503 .

byron — Reply to this email directly or view it on GitHub.

bwallace commented 10 years ago

will do thanks

On Fri, Nov 15, 2013 at 12:05 PM, George Dietz notifications@github.comwrote:

already done, care to remove this issue then?

On Nov 15, 2013, at 12:00 PM, byron wallace notifications@github.com wrote:

ok -- just make it 1 then?

On Fri, Nov 15, 2013 at 11:54 AM, George Dietz notifications@github.comwrote:

I disagree with this suggested enhancement. The error message we get when trying to use nAGQ = 3 (i.e. nAGQ != 1) is:

Error in updateGlmerDevfun(devfun, glmod$reTrms, nAGQ = nAGQ) : nAGQ > 1 is only available for models with a single, scalar random-effects term

Tom claims he doesn’t known what this means. Thus in lieu of a tom understanding, its foolhardy to try anything else.

On Nov 14, 2013, at 7:56 AM, byron wallace notifications@github.com wrote:

In bivariate.r, line 52 there is a call out to the bivariate package including this parameter: nAGQ = 3. This is the number of integration points. This should be a parameter to the method (so ultimately added to the wrapper method). The default should still be 3, I think.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub< https://github.com/bwallace/OpenMeta-analyst-/issues/243#issuecomment-28584503>

.

byron — Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHubhttps://github.com/bwallace/OpenMeta-analyst-/issues/243#issuecomment-28585493 .

byron

idahabreh commented 10 years ago

the default should not be 1 -- except as a temporary solution...

there are cases where having one integration point will produce wrong results - not common but it happens

George showed me what the problem is -- because of changes in the the generalized linear model packages in R (not under our control), he had to use glmer instead of lmer.

glmer does not support multiple integration points when there are 2 or more random effects (this probably tells you something about free software...)

On Fri, Nov 15, 2013 at 12:05 PM, George Dietz notifications@github.comwrote:

already done, care to remove this issue then?

On Nov 15, 2013, at 12:00 PM, byron wallace notifications@github.com wrote:

ok -- just make it 1 then?

On Fri, Nov 15, 2013 at 11:54 AM, George Dietz notifications@github.comwrote:

I disagree with this suggested enhancement. The error message we get when trying to use nAGQ = 3 (i.e. nAGQ != 1) is:

Error in updateGlmerDevfun(devfun, glmod$reTrms, nAGQ = nAGQ) : nAGQ > 1 is only available for models with a single, scalar random-effects term

Tom claims he doesn’t known what this means. Thus in lieu of a tom understanding, its foolhardy to try anything else.

On Nov 14, 2013, at 7:56 AM, byron wallace notifications@github.com wrote:

In bivariate.r, line 52 there is a call out to the bivariate package including this parameter: nAGQ = 3. This is the number of integration points. This should be a parameter to the method (so ultimately added to the wrapper method). The default should still be 3, I think.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub< https://github.com/bwallace/OpenMeta-analyst-/issues/243#issuecomment-28584503>

.

byron — Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHubhttps://github.com/bwallace/OpenMeta-analyst-/issues/243#issuecomment-28585493 .

Issa Dahabreh