# Publication bias: Egger's Test significant depending on subset. Why doesTrim and Fill not add studies?

For my final thesis I performed a meta-analysis with different subsets (subgroup analysis). Although I systematically searched the literature (I was brutally sophisticated), depending which subset I analyze, the Egger's Test is significant or not, indicating publicaton bias or not.

However, on the undifferentiated subset although the Egger's Test is significant, the Trim and Fill method does not add any studies to the sample.

Trim and Fill method is based on:

Duval, S., & Tweedie, R. 2000a. A Nonparametric “ Trim and Fill ” Method of Accounting for Publication Bias in Meta-Analysis. Journal of the American Statistical Association, 95(449): 89–98.

Duval, S., & Tweedie, R. 2000b. Trim and Fill: A Simple Funnel-Plot-Based Method. Biometrics, 56(June): 455–463.

I have several questions:

1) How to interpret that depending on the subset analyzed, the Egger's test is significant or not (I cannot derive a pattern or something similar).

2) Why does Trim and Fill not add studies, although Egger's Test is significant?

I will now add some pictures of the results, so you can better follow my thoughts. Thanks for any help.

• Welcome to our site! You might want to add a citation for the "trim and fill" method Sep 23, 2016 at 13:18
• One possible explanation is that Egger's method is parametric and Duval and Tweedie's is non-parametric. Sep 23, 2016 at 15:11
• @mdewey can u explain that a bit more sophisticatedly? I am no expert.
– Feal
Sep 23, 2016 at 15:41
• Please be aware that there are multiple papers recommending that you should not use Egger's test and the trim and fill method in the presence of between-study heterogeneity. Since your I^2 are very high, you have to be very carful in interpreting the results of these methods. Some references: http://www.ncbi.nlm.nih.gov/pubmed/12820277 and http://www.ncbi.nlm.nih.gov/pubmed/17420491 Sep 24, 2016 at 9:07
• @mdewey ha, thats what i've found out in the meantime as well. Thanks for ur patience & help.
– Feal
Sep 24, 2016 at 9:51

There is no particular reason to expect all the methods of looking for small study bias to agree. Song and colleagues here compared Egger, Begg, and Duval and Tweedie and showed that they did not always agree, see their Figure 1 for a quick insight.

• Well, still i am a bit confused how to interpret the results in the light of what these tests are telling me? Would you mind to give me a hint?
– Feal
Sep 23, 2016 at 16:10
• That must depend on (a) what your hypothesis about the small study effects was (b) what the actual shape of the funnel plot is like (c) what additional information you might have about the studies involved (d) perhaps something else which I have not thought of. Sep 23, 2016 at 21:08