PSY 1950 Meta-analysis December 3, 2008
Definition “the analysis of analyses... the statistical analysis of a large collection of analysis results from individual studies for the purpose of integrating the findings. It connotes a rigorous alternative to the casual, narrative discussions of research studies which typify our attempts to make sense of the rapidly expanding research literature.” –Glass (1976) “Mega-silliness” –Eysenck (1977)
History Pre-history –Pearson (1904), Fisher (1948), Cochran’s (1955) The Great Debate –1952: Eysenck concluded that psychotherapy was bunk –20 years of research did not settle debate –1978: Glass & Smith statistically aggregated findings from 375 studies, concluding that psychotherapy works Necessity is the mother of invention –Psychology abounds!
Rationale Meta-analyses avoid the limitations of qualitative/narrative/traditional reviews: –Weak effects overloooked Meta-analyses are more powerful –Differences between studies over-interpreted In meta-analysis, heterogeneity assessed statistically –Moderating variables overestimated or overlooked In meta-analysis moderators assessed statistically –Limited, subjective sampling of studies In meta-analysis, exhaustive search and defined inclusion/exclusion criteria –Overwhelmed by large database Meta-analyses can summarize hundreds of effects –Subjective assessment In meta-analysis, any subjectivity is discernible
Example: Finding Weak Effects Cooper, H. M., & Rosenthal, R. (1980). Statistical versus traditional procedures for summarizing research findings. Psychological Bulletin, 87, –32 grad students, 9 faculty members –Randomly assigned to statistical and traditional review technique –Given 7 studies that examined sex differences in persistence For 2 studies, females more persistent than males (ps =.005,.001) For other studies, no significant difference –Does evidence presented support the conclusions that females are more persistent?
actual p =.016
Example: Assessing Moderators Statistically
Criticisms Weak –Apples and oranges –Flat Earth society –Garbage in, garbage out –File-drawer problem Strong –Post-hoc
Apples and Oranges Critique –Meta-analyses add together apples and oranges Response –Glass: “in the study of fruit, nothing else is sensible” –Analogy with single experiments –Empirical question resolved through examination of moderating variables
Flat Earth Society Critique –Cronbach: "...some of our colleagues are beginning to sound like a kind of Flat Earth Society. They tell us that the world is essentially simple: most social phenomena are adequately described by linear relations; one- parameter scaling can discover coherent variables independent of culture and population; and inconsistences among studies of the same kind will vanish if we but amalgamate a sufficient number of studies.... The Flat Earth folk seek to bury any complex hypothesis with an empirical bulldozer.” Response –Code and analyze moderating variables
Garbage In, Garbage Out Critique –The inclusion of flawed studies “dirties” the database, obscures the truth, and invalidates meta-analytic conclusions Response –Glass: “I remain staunchly committed to the idea that meta-analyses must deal with all studies, good bad and indifferent, and that their results are only properly understood in the context of each other, not after having been censored by some a priori set of prejudices.” –Empirical question: Study quality (or better yet, related variables) can be coded and analyzed as moderators
File-drawer Problem Critique –Meta-analytic database is biased sampling of studies –Significant findings are more likely to be published than nonsignificant findings Response –Less bias than narrative reviews –File-drawer analyses (e.g., funnel plots) can empirically address the presence and influence of missing studies
Post-hoc Criticism –By definition, meta-analysis is a post-hoc endeavor, i.e., an observational study Moderating variables may be confounded, sometimes extremely so Effects may be correlational Response –Confounding may be interested in its own right –Statistical control –Hypothesis generation versus hypothesis testing
Steps of a Meta-analysis Define question Search literature Determine inclusion/exclusion criteria Code moderating variables Analyze data This is an iterative process!
Defining Meta-analytic Question Interestingness –Establish presence of effect –Determine magnitude of effect –Resolve differences in literature –Test competing theories e.g., psychotherapy, imagery v1
Inclusion/Exclusion Criteria Theoretical considerations –Scope/generalizability –Quality Practical considerations –Power –Missing data –Time
Studies were included if they –had written published or unpublished reports in English available by March 1, 2008 –presented original data from between-participants, within-participants (i.e., single-group pretest-posttest, or PP), or mixed design (i.e., independent-groups pretest-posttest, or IG-PP) experiments or quasi- experiments –objectively, quantitatively evaluated performance on at least one cognitive task as a function of meditative experience or state Studies were excluded if they –used a psychopathologically or neurologically disordered population –confounded meditation with other mental training (e.g., education), maturation, or practice and used measures susceptible to such confounding (e.g., academic achievement test) –did not report data on or contained data that allow estimation of participants’ age or meditative experience –did not contain basic methodological information (e.g. type of task administered)
Literature Search Types of searches –Keyword –Ancestor –Descendent Available Resources –Electronic e.g., PsychInfo, SCI, Google scholar –Physical Conference proceedings Bibliographies Key journals –Mental Experts in the field
Harvard’s Electronic Resources SSCI/SCI (Social/Science Citations Index) – PsychInfo – Google Scholar – Harvard – HOLLIS – Interlibrary Loan – Dissertations (Proquest) –
Coding What to code –Anything possibly interesting e.g., control group/condition, participant variables –Anything possibly confounding e.g., publication year, journal impact factor –How you coded effect sizes How to code –Using explicit coding scheme –Set measurement scale –Multiple coders –Calculate reliability
Analysis Calculate effect size Weight effect size Describe Infer –Univariate analyses –Multivariate analyses
Calculating Effect Size Only one ES per construct per study –Balance between dependency and thoroughness Typically d or r Can be calculated in lots of way (from raw data to graphs) Effect size calculator
Weighting Effect Size Why weight? –Studies vary significantly in size –Studies with large n have more reliable effect sizes than studies with small n How weight? –Simple approach: weight by sample size –Better approach: weight by precision What is precision weighting? –Each effect size has associated SE –Hedges showed that best meta-analytic estimate of precision is weight by inverse sampling variance
Describing Distribution Central tendency Spread Shape
Inferencial Statistics Select a model –Fixed effects –Random effects Univariate analyses –Analogous to one-way ANOVA –Examine how much variation in effect sizes is explained by one (categorical) variable Multivariate analyses –Analogous to multivariate regression –Examine how much variation in effect sizes is explained by set of (categorical or continuous) variables –Examine how much unique variation in effect sizes is explained by one (categorical or continuous) variable