14: Meta-Analysis

Quantitative Research Synthesis

Dean Adams, Iowa State University

Outline

For further information see: Cooper and Hedges (1994). Handbook of Research Synthesis. Hedges and Olkin (2000). Statistical Methods for Meta-Analysis. Rosenberg, Adams, and Gurevitch (2000). MetaWin: Statistical Software for Meta-Analysis. vsn. 2.

Synthesizing Prior Research

Synthesizing Prior Research

Quant. Res. Synthesis: A Brief History

Preliminary Concepts

* NOTE: unpublished studies that can be obtained from authors can also be included

1: Vote Counting

1: Vote Counting

2: Combining Probabilities

2: Combining Probabilities

NOTE: often called omnibus tests (only depend on exact probabilities of each study)

2: Combining Probabilities: Some Methods

\[\small\alpha={1-(1-\alpha^*)^{1/n}}\]

\[\small{P}= -2\sum{log{(p_i)}}\]

NOTE: see similarity to LRT tests

\[\small{Z}=\sum{Z(p_i)}/\sqrt{n}\]

\[\small{P_E}=(\sum{p_i})^n/{n!}\]

Sum of logs: Example

\[\small{p_i}=(0.06; 0.02; 0.035; 0.001; 0.24)\]

\[\small{log(p_i)}=(-1.22; -1.77; -1.46; -3; -0.62)\]

3: Meta-Analysis

Effect Sizes

NOTE: By this point in the semester, you should be well-familiar with the concept of an effect size!

Effect Sizes from \(\small\overline{X}\) and \(\small\sigma\)

Effect Sizes from \(\small{2\times2}\) Tables

Effect Sizes from Correlations (and other Statistics)

Meta-Analytic Models

NOTE: All models are actually special cases of same model!

Meta-Analysis: No Structure

Meta-Analysis: No Structure

Meta-Analysis: Categorical Structure

Meta-Analysis: Categorical Structure

Meta-Analysis: Continuous Structure

\[\small{\beta_1}=\frac{\sum{w_iX_iE_i}- \frac{\sum{w_iX_i}\sum{w_iE_i}}{\sum{w_i}} }{\sum{w_iX_i}-\frac{(\sum{w_iX_i})^2}{\sum{w_i}}}\]

\[\small{\beta_0}=\frac{\sum{w_iE_i}-\beta_1\sum{w_iX_i}}{\sum{w_i}}\]

Meta-Analytic Models: Comments

\[\small{\overline{\overline{E}}}=\frac{\sum{w_iE_i}}{\sum{w_i}}\] \[\small{Q_T}=\sum{w_i}(E_i-\overline{\overline{E}})^2\]

Meta-Analytic Models: Comments

\[\small{\overline{\overline{E}}}=\frac{\sum{w_iE_i}}{\sum{w_i}}\] \[\small{Q_T}=\sum{w_i}(E_i-\overline{\overline{E}})^2\]

Meta-Analysis via GLS

\[\small\mathbf{E}=\mathbf{X{\hat{\beta}}+\epsilon}\]

Where variation in effect sizes (\(\small\mathbf{E}\)) is explained by some statistical design as found in \(\small\mathbf{X}\).

\[\tiny\mathbf{W}= \left( \begin{array}{ccc} w_1 & 0 & 0 & 0 \\ 0 & w_2 & 0 & 0 \\ \vdots & \vdots & \ddots & \vdots \\0 & 0 & 0 & w_n \end{array} \right) \]

Meta-Analysis via GLS

\[\small\mathbf{E}=\mathbf{X{\hat{\beta}}+\epsilon}\]

Where variation in effect sizes (\(\small\mathbf{E}\)) is explained by some statistical design as found in \(\small\mathbf{X}\).

\[\tiny\mathbf{W}= \left( \begin{array}{ccc} w_1 & 0 & 0 & 0 \\ 0 & w_2 & 0 & 0 \\ \vdots & \vdots & \ddots & \vdots \\0 & 0 & 0 & w_n \end{array} \right) \]

Meta-Analysis: Fixed vs. Random Effects

Meta-Analysis: Example

Meta-Analysis: Example (Cont.)

Publication Bias

-Can be assessed in a number of ways:

-Funnel Plot: plot effect size vs. sample size: should be funnel shaped (larger variance with smaller n). If overabundance of extreme values (for given n) with lack of data ‘in’ funnel, might be publication bias

Publication Bias (Cont.)

\[\small{N_R}=\frac{\sum({Z_p}_i)^2}{Z^2_\alpha}-n\]

Cumulative Meta-Analysis

Evaluating Significance

Phylogenetic Meta-Analysis

Meta-Analysis: Conclusions