[R-meta] meta-regression: categorical

[Michael Dewey]

Dear Roberto

I suspect the answer is that your original factor variable has levels 
which do not appear in your data-set so they are dropped when the model 
matrix is formed.

If you do factor(already_existing_factor) it silently drops the levels. 
This is rather hidden in the documentation as the second paragraph of Value

Michael

On 06/01/2020 09:27, P. Roberto Bakker wrote:
> Hi Everybody,
> 
> I have a new question about meta-regression.
> My database consists only of factors (no characters), so I suppose I do not
> need 'factor()' in 'mods', i.e. 'mods = ~ factor-var'. I also tried 'mods =
> ~ factor(factor-var)' and I receive the same results as expected.
> Only, there is one difference, 'mods = ~ factor-var' gives a warning
> message:
> *"In rma(measure = "SMCC", yi = yi, vi = vi, data = datsub, digits = 2,  :*
> *  Redundant predictors dropped from the model."*
> Whereas 'mods = ~ factor(factor-var)' does not give this message. What is
> the reason for this?
> 
> So, my question is not why I receive this message, but why the difference:
> with/out warning message.
> 
> Best regards,
> Roberto
> PS the same happens with a character variable: i.e. 'mods = ~ chr-var' as
> well as 'mods = ~ factor(chr-var)'.
> 
> Op do 26 apr. 2018 om 07:42 schreef P. Roberto Bakker <
> robertobakker using gmail.com>:
> 
>> Hi Wolfgang,
>>
>> Thank you for your information and explanation.
>>
>> Bw Roberto
>>
>> 2018-04-24 12:25 GMT+02:00 Viechtbauer, Wolfgang (SP) <
>> wolfgang.viechtbauer using maastrichtuniversity.nl>:
>>
>>> Hi Roberto,
>>>
>>> If you put a character variable (or a factor) into a formula, it is
>>> automatically dummy coded. Actually, the type of coding depends on:
>>>
>>> options("contrasts")
>>>
>>> But the default for that is:
>>>
>>>          unordered           ordered
>>> "contr.treatment"      "contr.poly"
>>>
>>> And help(contr.treatment) explains what kind of coding this is, namely
>>> the 'usual' dummy coding where one level is the 'reference' level (and
>>> whose dummy gets omitted).
>>>
>>> Best,
>>> Wolfgang
>>>
>>> -----Original Message-----
>>> From: R-sig-meta-analysis [mailto:
>>> r-sig-meta-analysis-bounces using r-project.org] On Behalf Of P. Roberto Bakker
>>> Sent: Tuesday, 24 April, 2018 11:14
>>> To: r-sig-meta-analysis using r-project.org
>>> Subject: [R-meta] meta-regression: categorical
>>>
>>> Hi everybody,
>>>
>>> In:
>>> http://www.metafor-project.org/doku.php/analyses:vanhouwelingen2002
>>> 'allocation' is put in syntax without dummy coding
>>>
>>> rma <http://finzi.psych.upenn.edu/library/metafor/html/rma.uni.html>(yi,
>>> vi, mods = ~ alloc, data
>>> <http://stat.ethz.ch/R-manual/R-devel/library/utils/html/data.html>=dat,
>>> method="ML")
>>>
>>> If I am not wrong, 'allocation' is a catergorical variabel, and I suppose
>>> the R handles the coding of dummy automatically.
>>> Correct?
>>>
>>> I found this information in:
>>> https://cran.r-project.org/web/packages/metafor/metafor.pdf
>>> *Categorical moderator variables can be included in the model via the mods
>>> argument in the same way that appropriately (dummy) coded categorical
>>> independent variables can be included in linear models. One can either do
>>> the dummy coding manually or use a model formula together with the factor
>>> function to let R handle the coding automatically.   *
>>>
>>> *Best wishes and thank you in advance*
>>> *Roberto*
>>>
>>
>>
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-meta-analysis mailing list
> R-sig-meta-analysis using r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis
> 

-- 
Michael
http://www.dewey.myzen.co.uk/home.html