# [R-meta] vi versus sei in rma() give different results

P. Roberto Bakker robertobakker at gmail.com
Wed Oct 18 22:44:59 CEST 2017

```Hi,

I understood from the metafor documentation you can either use vi
(variance) or sei (standard error (=sqrt of variance) in rma()..
However, with vi and I get other results than with sei. I would expect the
same results should be the same.
Can somebody help me understand this, or am I doing something wrong?

Roberto
PS in the meta-analysis literature I read that se = sqrt of variance. But I
learned from statistics that SD = sqrt of variance. Can somebody explain me
this?

> re
# A tibble: 10 × 6
yi       sei         vi `inverse varience` `weighed hedge`
X__1
<dbl>     <dbl>      <dbl>              <dbl>           <dbl>
<dbl>
1   1.91498797 0.4231468 0.17905321           5.584932      10.6950776
1.91498797
2  -0.25848358 0.2615776 0.06842283          14.615005      -3.7777390
-0.25848358
3   0.06543005 0.3204569 0.10269263           9.737797       0.6371445
0.06543005
4   0.77363080 0.3290877 0.10829873           9.233718       7.1434890
0.77363080
5  -0.79554100 0.3341517 0.11165733           8.955973      -7.1248437
-0.79554100
6   0.29374938 0.2387312 0.05699258          17.546145       5.1541691
0.29374938
7   0.94409401 0.2527629 0.06388910          15.652122      14.7770750
0.94409401
8  -0.56731857 0.2423818 0.05874893          17.021588      -9.6566629
-0.56731857
9   0.18910949 0.1766278 0.03119739          32.053967       6.0617093
0.18910949
10  0.38103574 0.1713235 0.02935176          34.069510      12.9817010
0.38103574
>

WITH SEI
*> reis <- rma(measure="SMD", yi, sei, data=re, method="FE", digits=2)*
*> reis*

*Fixed-Effects Model (k = 10)*

*Test for Heterogeneity: *
*Q(df = 9) = 16.54, p-val = 0.06*

*Model Results:*

*estimate    se  zval  pval  ci.lb <http://ci.lb>  ci.ub   *
*    0.25  0.16  1.54  0.12  -0.07   0.56   *

*---*
*Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 *

*WITH VI*

*> res <- rma(measure="SMD", yi, vi, data=re, method="FE", digits=2)>
resFixed-Effects Model (k = 10)Test for Heterogeneity: Q(df = 9) = 51.46,
p-val < .01Model Results:estimate    se  zval  pval  ci.lb <http://ci.lb>
ci.ub        0.22  0.08  2.88  <.01   0.07   0.38  **---Signif. codes:  0
‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 *

[[alternative HTML version deleted]]

```