[R] Problems using hetcor (polycor)

[Mark Difford]

Hi Birgitle,

>> It seems than, that it is not possible to use all variables without
>> somehow 
>> imputing missing values.

It depends on what you are after. You can use the full data set if you set
std.err=F and pd=F. Then exclude the columns that cause it to falter and
redo with SEs turned on. You have the correlations; all you lack are SEs for
ca. 5 columns.

And I haven't tried your revised classification; that's for you to do.

Regards, Mark.


Birgitle wrote:
> 
> Many Thanks Mark for your answer.
> 
> It seems than, that it is not possible to use all variables without
> somehow imputing missing values.
> 
> But I will try which variables I can finally use.
> 
> Many thanks again.
> 
> B.
> 
> 
> Mark Difford wrote:
>> 
>> Hi Birgitle,
>> 
>> It seems to be failing on those columns that have just a single "entry"
>> (i.e = 1, with the rest as 0; having just 1, an <NA>, and then 0s gets
>> you through). And there are other reasons for failure (in the call to get
>> a positive definite matrix).
>> 
>> The main problem lies in the calculation of standard errors for some
>> combinations. You can get a result for all columns by turning this off.
>> You can get standard errors for all the columns up to 60 by omitting
>> columns 12, 23, and 44.  You need to work out the rest yourself by
>> "columning" forwards till you get a failure, then drop that column from
>> the index. There probably isn't enough information to calculate SEs for
>> the columns that cause an error.
>> 
>> ## This works with default settings but without columns 12, 23 and 44
>> hetcor(TestPart[,c(1:11,13:22,24:43,45)], pd=T, std.err=T,
>> use="complete.obs")
>> 
>> ## The first fails; the second works
>> hetcor(TestPart[,c(1:11,13:22,24:43,45:60)], pd=T, std.err=F)
>> hetcor(TestPart[,c(1:72)], pd=F, std.err=F)
>> 
>> HTH, Mark.
>> 
>> 
> 
> 

-- 
View this message in context: http://www.nabble.com/Problems-using-hetcor-%28polycor%29-tp18867343p18870999.html
Sent from the R help mailing list archive at Nabble.com.