[R-meta] Wald_test error

[Cátia Ferreira De Oliveira]

Dear James,

Apologies for not adding the mailing list to the email.
I am planning many contrasts, they could be accomplished by subgrouping the
data but not sure if that's a good approach. What I am aiming to accomplish
is to determine whether there are
a) any differences between the TD group and DD/DLD groups for grammar,
vocabulary and phonology;
b) comparing whether the effect for grammar, phonology and vocabulary == 0;
c) pairwise comparisons of the effect for grammar, phonology and vocabulary
to see if they are different;
d) check whether the correlation is different from 0 for the reference
level (TD) for each component - grammar, vocabulary, phonology (not sure if
I would need to run a different model for this one).

The reference level is the TD group, there are three groups (TD, DD and
DLD) and three components (grammar, vocabulary and phonology).


*component <- robu(formula = yi ~ 0 + Component + Group:Component, data =
df,*
*                       studynum = Study, var.eff.size = vi,*
*                       rho = .8, small = TRUE)*
*print(component)*


*RVE: Correlated Effects Model with Small-Sample Corrections*

*Model: yi ~ 0 + Component + Group:Component*



*Number of studies = 34 *
*Number of outcomes = 305 (min = 1 , mean = 8.97 , median = 4 , max = 48 )*
*Rho = 0.8 *
*I.sq = 47.09732 **Tau.sq = 0.02458387 *


*                               Estimate StdErr t-value   dfs P(|t|>) 95%
CI.L 95% CI.U Sig*
*1             ComponentGrammar  0.05897 0.0456   1.293 11.45  0.2217
 -0.0410   0.1589    *
*2           ComponentPhonology -0.00383 0.0356  -0.108  8.05  0.9169
 -0.0857   0.0781    *
*3          ComponentVocabulary  0.09662 0.0592   1.631  5.35  0.1600
 -0.0527   0.2460    *
*4     ComponentGrammar.GroupDD  0.03277 0.0590   0.555  1.05  0.6740
 -0.6412   0.7068    *
*5   ComponentPhonology.GroupDD  0.01557 0.0929   0.168  5.60  0.8728
 -0.2159   0.2470    *
*6  ComponentVocabulary.GroupDD -0.16516 0.0592  -2.788  5.35  0.0358
 -0.3145  -0.0158  ***
*7    ComponentGrammar.GroupDLD -0.08097 0.0819  -0.989 14.99  0.3386
 -0.2556   0.0936    *
*8  ComponentPhonology.GroupDLD  0.35854 0.1760   2.037  5.77  0.0897
 -0.0763   0.7934   **
*9 ComponentVocabulary.GroupDLD -0.10642 0.0692  -1.537  9.94  0.1555
 -0.2608   0.0480    *

*---*
*Signif. codes: < .01 *** < .05 ** < .10 **
*---*
*Note: If df < 4, do not trust the results*



Thank you! I am a bit new to using the constraints.

Best wishes,

Catia

On Sun, 15 Aug 2021 at 03:38, James Pustejovsky <jepusto using gmail.com> wrote:

> Please keep the list cc'd. Responses below.
>
> James
>
> On Fri, Aug 13, 2021 at 9:56 AM Cátia Ferreira De Oliveira <
> cmfo500 using york.ac.uk> wrote:
>
>> Dear James,
>>
>> Thank you for your reply! I wasn't able to find any examples similar to
>> mine, so could you give me an idea of how one would go about doing the
>> constraints when you want to test for an overall group effect and there's
>> an interaction term?
>>
>> group <- robu(formula = yi ~ 0 + Group + Component:Group, data = Data,
>> studynum = Study, var.eff.size = vi,
>>                        rho = .8, small = TRUE)
>> print(group)
>>
>
> To answer this question, we need to know what the null hypothesis of
> interest is. In the model as you've specified it, the definition of the
> group effects depends on how you specify the contrasts for the Component
> term. As a result, it's not clear what the main effects of the Group term
> mean. Could state in words what hypothesis you're trying to test?
>
>
>>
>> Also, just to confirm, if there is only one predictor with three levels
>> (yi ~ 0 + Variable), would the constraints be the following:
>>
>> Wald_test(model, constraints = matrix(c(1,0,0,0,1,0,0,0,1),3,3), vcov =
>> "CR2")
>>
>> Again, we need to know what the null hypothesis of interest is. Using a
> diagonal constraint matrix (as you've specified) will test the null that
> the average effect size is equal to zero for each of three levels of
> Variable. That might be of interest, or perhaps you instead want to test
> whether the average effect sizes are *identical* across the three levels of
> Variable (but not necessarily all zero). For the latter null, you would
> instead use
>
> constraints = constrain_equal(1:3)
>
> or
>
> constraints = constrain_equal("Variable", reg_ex = TRUE)
>
>

-- 
Cátia Margarida Ferreira de Oliveira
Psychology PhD Student
Department of Psychology, Room B214
University of York, YO10 5DD
pronouns: she, her

	[[alternative HTML version deleted]]