[R-meta] Two questions about the subgroup analysis

Viechtbauer Wolfgang (SP) wolfgang.viechtbauer at maastrichtuniversity.nl
Mon Feb 5 10:54:14 CET 2018


With 'total variance', do you mean the Q-statistic for the test of heterogeneity? No, that is the same.

However, only for fixed-effects models do we find that:

Q(Between groups) + Q(Within groups) = Q(Total),

which we can also see in the output below: 12.48 + 47.69 =~ 60.18 (the slight discrepancy is due to rounding). In random/mixed-effects models, this decomposition doesn't work anymore, since Q(Between groups) is computed in a different way (the statistic incorporates the amount of residual heterogeneity).

Best,
Wolfgang

>-----Original Message-----
>From: 전경남 [mailto:fstyle71 at naver.com]
>Sent: Monday, 05 February, 2018 1:43
>To: Viechtbauer Wolfgang (SP); r-sig-meta-analysis at r-project.org
>Subject: RE: [R-meta] Two questions about the subgroup analysis
>
>Dear Wolfgang,
>
>Thank you so much for your help!
>Using the command you provided, I can get the p-values as I need.
>I have one more question.
>If only between-group variance is different under boht models,
>will I get the different total variance under different models?
>
>Kyungnam
>
>-----Original Message-----
>From: "Viechtbauer Wolfgang
>(SP)"<wolfgang.viechtbauer at maastrichtuniversity.nl>
>To: "전경남"<fstyle71 at naver.com>; "r-sig-meta-analysis at r-project.org"<r-
>sig-meta-analysis at r-project.org>;
>Cc:
>Sent: 2018-02-04 (일) 21:12:06
>Subject: RE: [R-meta] Two questions about the subgroup analysis
>
>Dear Kyungnam,
>
>1) The CI for each subgroup tells you whether the effect for that
>subgroup is significantly different from 0. If 0 is not part of the 95%
>CI, then you know that the effect is significantly different from 0 at
>alpha = .05 (two-sided). If you want the p-values for the tests, then you
>can find them with:
>
>res$pval.fixed.w
>res$pval.random.w
>
>for the fixed- and random-effects models, respectively, and where 'res'
>is the fitted model object.
>
>2) The "Within groups" test is the same for the fixed- and random-effects
>model. Hence, it is only shown once.
>
>Best,
>Wolfgang
>
>--
>Wolfgang Viechtbauer, Ph.D., Statistician | Department of Psychiatry and
>Neuropsychology | Maastricht University | P.O. Box 616 (VIJV1) | 6200 MD
>Maastricht, The Netherlands | +31 (43) 388-4170 | http://www.wvbauer.com
>
>>-----Original Message-----
>>From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces at r-
>>project.org] On Behalf Of ???
>>Sent: Saturday, 03 February, 2018 3:45
>>To: r-sig-meta-analysis at r-project.org
>>Subject: [R-meta] Two questions about the subgroup analysis
>>
>>Dear all,
>>
>>I performed a subgropu analysis using the meta package.
>>The following are the results of the analysis.
>>I have two questions about the results.
>>
>>1. Is there any option that I can see the results of significant
>>testsings
>>about whether each effect size estimates for the two groups are zero or
>>not ?
>>
>>2. Why could I get the results only for between groups under the random
>>model?
>>As you see, under the fixed effect model,
>>information about both between groups and within groups are provided.
>>
>>Thank you for your help in advance!
>>
>>Kyungnam Jeon
>>
>>----- The results of analysis ------
>>
>>Number of studies combined: k = 15
>>
>>                        SMD            95%-CI    z  p-value
>>Fixed effect model   0.0993 [ 0.0317; 0.1670] 2.88   0.0040
>>Random effects model 0.1026 [-0.0407; 0.2458] 1.40   0.1605
>>Quantifying heterogeneity:  tau^2 = 0.0595; H = 2.07 [1.62; 2.66]; I^2 =
>>76.7% [61.8%; 85.8%]
>>
>>Test of heterogeneity:
>>     Q d.f.  p-value
>> 60.18   14 < 0.0001
>>
>>Results for subgroups (fixed effect model):
>>                 k     SMD            95%-CI     Q    tau^2   I^2
>>level = ele     10  0.1959 [ 0.1097; 0.2822] 40.71   0.0696 77.9%
>>level = second   5 -0.0544 [-0.1633; 0.0544]  6.99   0.0117 42.8%
>>
>>Test for subgroup differences (fixed effect model):
>>                   Q d.f.  p-value
>>Between groups 12.48    1   0.0004
>>Within groups  47.69   13 < 0.0001
>>
>>Results for subgroups (random effects model):
>>                 k     SMD            95%-CI     Q    tau^2   I^2
>>level = ele     10  0.1873 [-0.0018; 0.3764] 40.71   0.0696 77.9%
>>level = second   5 -0.0538 [-0.1990; 0.0915]  6.99   0.0117 42.8%
>>
>>Test for subgroup differences (random effects model):
>>
>>                    Q d.f.  p-value
>>Between groups   3.93    1   0.0475
>>
>>Details on meta-analytical method:
>>- Inverse variance method
>>- DerSimonian-Laird estimator for tau^2
>>- Hedges' g (bias corrected standardised mean difference)


More information about the R-sig-meta-analysis mailing list