Meta-analysis

A meta-analysis is a summary of primary studies on metadata that works with quantitative, statistical methods. The term was coined by psychologist Gene V Glass ( born 1940 ) introduced in his article "Primary, Secondary and Meta - Analysis of Research " in 1976. It defines meta-analysis as [ ... ] analysis of Analyses. I use it to refer to the statistical analysis of a large collection on analysis results from individual studies for the purpose of integra ting the findings. ( German: " [ ... ] analysis of analyzes that I mean the statistical analysis of a large collection of analysis results of several individual studies that are to be merged result.. "). Was conducted the first meta-analysis, however, already in 1904 by Karl Pearson, who wanted to increase the statistical power of studies with few subjects by grouping.

In contrast to the method of meta-analysis are qualitative method in which attempts by way of subjective assessment, to be drawn from the contents of the primary studies conclusions. When using qualitative methods erroneous conclusions or evaluations are much more likely than in meta-analyzes.

Areas of application and reasons

Meta-analysis allows for the aggregation of various investigations into a scientific research area. The individual empirical results of content homogeneous primary studies are summarized. The goal is an effect size estimate. It will be investigated whether there is an effect, and the size of this is.

Meta-analyzes are often used esp. in medicine, psychology and social research. There they help with two problems:

Method

A meta-analysis includes all elements of the social science research process, as they are also taken with a primary research ( Cooper 1982, Schnell et al 1995. ).

The process of meta-analysis is similar to the narrative review, the structure presents the relevant literature on a scientific topic and provide you with critical comments. Criticized the subjectivity of the selection of studies. Here, the meta-analysis achieved a greater objectivity, by establishing criteria for the selection of primary studies for meta-analysis, by itself, however, reduces the possible number of studies that can be included in a meta-analysis.

The summary of various studies to a scientific research field is only meaningful when the effect sizes of the individual studies are estimates of a common population effect size. It is a homogeneity test is necessary.

Homogeneity tests assume a uniform effect size Δ (read: delta) from. Δ is a universal Effektgrößenmaß and corresponds to the bivariate product moment correlation. It prefers, because different statistical parameters (eg, r, t, F) can be transformed into Δ.

The study-specific effect sizes are then checked by a significance test for homogeneity. Are the effect sizes homogeneous, the average Δ value can be calculated, it corresponds to the estimate of the population effect size and can be tested for significance and classified with respect to its size.

If there are heterogeneous effect sizes, so you can apply strategies that share the studies with heterogeneous effect sizes into homogeneous subgroups. Subsequently, the influence of moderator variables on the heterogeneity should be determined, this happens variance or cluster analysis. If there are no direct assumptions about the effect of moderator variables, a correlation between the moderator variable and the study-specific effect sizes can be calculated. The level of correlation here describes the influence of moderator variables on the heterogeneity of the effect sizes.

Since investigation reports are often incomplete and sometimes only reported significant or non- significant results, there are methods that make it possible to also use these studies meta-analysis (eg, counting, sign test, binomial test and calculation of the exact probability of error, it results in the Stouffer test statistic ). Using the 'fail -safe N "( Rosenthal ) can be calculate in the presence of a significant overall test, as many studies with a mean effect size of zero would still be present in addition to ensure that the overall test is not significant.

Criticism

Garbage -in- garbage -out problem: It criticizes the results of a meta-analysis are less valid because any investigation is received, regardless of their methodological quality in the meta-analysis. However, the influence of methodological quality of the studies are controlled on the results of the meta-analysis by evaluation criteria are used, the basis of which evaluate the results of an investigation and their effect size can be weighted so.

Apples and oranges problem: One criticism is that meta-analyzes combine studies with different Operationalisierungsvarianten. It is claimed that, especially in relation to the dependent variable operationalization must be homogeneous, since they all have to be indicators of the same construct. Otherwise, the studies refer to different criteria, a summary would not be useful.

Drawer problem: Often, only results will be published, confirm the hypotheses adopted or have significant results, while studies are not published with non-significant results ( publication bias ). This results in a distortion of the meta-analytic results, as they demonstrate the existence of an effect too often. Unpublished literature also referred to as gray literature. " Interestingly, the gray literature in the former GDR as not censored literature usually have a higher scientific value than the official, state-controlled literature. " Psychological Review " A list of recent theses and dissertations in psychology semi-annually resolved. " The problem of gray literature is get hold treated at Marylu C. Rosenthal ( 1994).

Problem of dependent measurements: This problem occurs when several (dependent) partial results have been collected on the same sample. As units of analysis of meta-analyzes but are individual studies and not sub-samples, only a result of an investigation into the meta-analysis must always be received with, because otherwise this investigation would receive greater weight than a study which deals only with a result in the meta-analysis. Either choose the most important result under the partial results, or by forming the arithmetic mean as an estimate of the overall result.