You are on page 1of 6

Methods

Grey Literature in
Meta-Analyses
Vicki S. Conn ▼ Jeffrey C. Valentine
Harris M. Cooper ▼ Marilyn J. Rantz

 Background: In meta-analysis, researchers combine the results of individual studies to manuscripts, non-English language
arrive at cumulative conclusions. Meta-analysts sometimes include “grey literature” in articles, and technical reports (Cook
their evidential base, which includes unpublished studies and studies published out- et al., 2001; Dickersin, 1994; Eysen-
side widely available journals. Because grey literature is a source of data that might bach, Tuische, & Diepgen, 2001;
not employ peer review, critics have questioned the validity of its data and the results McAuley, Pham, Tugwell, & Moher,
of meta-analyses that include it. 2000). Serious concerns have been

raised about the inclusion of grey lit-
Objective: To examine evidence regarding whether grey literature should be included in
erature. Although meta-analyses are
meta-analyses and strategies to manage grey literature in quantitative synthesis.
increasingly common in nursing, the
 Methods: This article reviews evidence on whether the results of studies published in methodological issue of inclusion of
peer-reviewed journals are representative of results from broader samplings of grey literature has not been addressed
research on a topic as a rationale for inclusion of grey literature. Strategies to enhance in nursing journals. This article exam-
access to grey literature are addressed. ines issues related to inclusion of grey
 Results: The most consistent and robust difference between published and grey litera- literature and strategies for handling
ture is that published research is more likely to contain results that are statistically sig- grey literature in meta-analyses.
nificant. Effect size estimates of published research are about one-third larger than
those of unpublished studies. Unfunded and small sample studies are less likely to be Brief Overview of
published.Yet, importantly, methodological rigor does not differ between published and Meta-Analysis
grey literature.

Meta-analysis research analyzes the
Conclusions: Meta-analyses that exclude grey literature likely (a) over-represent stud-
pooled results of several primary stud-
ies with statistically significant findings, (b) inflate effect size estimates, and (c) provide
ies to provide a quantitative review of
less precise effect size estimates than meta-analyses including grey literature. Meta- existing empirical evidence. Meta-
analyses should include grey literature to fully reflect the existing evidential base and analysts calculate an overall statistic
should assess the impact of methodological variations through moderator analysis. to estimate the magnitude of associa-
 Key Words: meta-analysis 䡠 research design tion between the variables being stud-
ied. Subsequent moderator analysis
examines differences in effect size
associated with variations between

M eta-analysis is a powerful tool


for summarizing research
(Cooper & Hedges, 1994). It can pro-
accepted source of research but meta-
analysts sometimes include so-called
grey literature. Grey literature refers
studies. For example, a recent meta-

vide a scientific evidence base for nurs- to studies with limited distribution
ing practice and guide future nursing (i.e., those not included in computer- Vicki S. Conn, PhD, RN, is Professor and
research (Conn & Armer, 1996). ized bibliographic retrieval systems), Associate Dean for Research, and Mari-
However, only the studies chosen for unpublished reports, dissertations, lyn J. Rantz, PhD, RN, FAAN, is Profess-
inclusion in a meta-analysis can deter- articles in obscure journals, some sor, School of Nursing; Jeffrey C. Valen-
tine, PhD, is Research Assistant Professor,
mine its results. online journals, conference abstracts, and Harris M. Cooper, PhD, is Professor,
Studies published in peer-reviewed policy documents, reports to funding Department of Psychological Sciences;
scientific journals are the most widely agencies, rejected or unsubmitted University of Missouri-Columbia.

256 Nursing Research July/August 2003 Vol 52, No 4


Nursing Research July/August 2003 Vol 52, No 4 Grey Literature in Meta-Analyses 257

analysis reported the overall effect size statistically significant, commonly bias against the null hypothesis as well
of interventions to increase physical referred to as bias against the null as any other bias that makes the
activity among aging adults and sig- hypothesis (Dickersin, Min, & Mein- results of published research different
nificantly larger effect sizes for stud- ert, 1992; Easterbrook, Berlin, from other research due to factors
ies with selected intervention compo- Gopalan, & Matthews, 1991; Hub- other than quality. For example, these
nents (e.g., self-monitoring) and for bard & Armstrong, 1997; Sterling, biases include the tendency for
studies with particular subject char- Rosenbaum, & Weinkam, 1995). For authors to submit, and editors to
acteristics (e.g., patients with specific example, Dickersin and Min (1993) accept for publication, research that is
chronic illnesses) (Conn, Valentine, reported an odds ratio (OR) for the consistent with previously published
& Cooper, 2002). Moderator analy- association between significant results findings (Cooper, DeNeve, &
sis is especially useful for nursing and publication of 6.15 (95% CI 2.24 Charleton, 1997). Failure to include
intervention research where varia- to 16.92). Bias in favor of significant unpublished studies compromises the
tions in interventions between studies findings is present even when quality validity and reliability of meta-analy-
are common. The selection of studies is controlled (Stern & Simes, 1997). sis when unpublished findings differ
for the quantitative synthesis is sig- Thus, even among randomized con- in some systematic way from pub-
nificant for the accuracy of the over- trolled trials, bias against the null lished findings (Dickersin, 1997).
all effect size estimate and for ensur- Unpublished and published
ing sufficient variability for research differs because investigators
moderator analysis. halt studies at preliminary stages
The immense potential benefits of when data do not favor the experi-
meta-analyses can be realized only The immense potential mental treatment (Dickersin, Chan,
when appropriate methods are Chalmers, Sacks, & Smith, 1987).
applied to the selection and manage- benefits of meta-analyses Unpublished research is more likely to
ment of primary studies. Most meta- can be realized only when have small samples, which may reflect
analyses exclude grey literature and intriguing pilot projects, difficult-to-
appropriate methods
research reports published in lan- recruit subjects, or highly innovative
guages other than English (Gregoire, are applied to the selection interventions (Chalmers et al., 1990;
Derderian, & Le Lorier, 1995; and management of Dickersin et al., 1987; Easterbrook et
McAuley et al., 2000). Sometimes the al., 1991; Thornton & Lee, 2000).
exclusion is a deliberate a priori deci- primary studies. Exclusion of these small studies from
sion perhaps using publication in meta-analysis is particularly unfortu-
widely disseminated journals as a nate since one of the method’s advan-
proxy measure for study quality. tages is its ability to summarize results
Often, grey literature is not retrieved across small samples.
because the researchers limit their Externally funded research is more
search strategies to computerized hypothesis remained strong (OR ⫽ likely to be published (Dickersin et al.,
databases (e.g., MEDLINE) that are 8.92, 95% CI 1.96 to 40.65). 1992; Dickersin & Min, 1993; Stern
unlikely to access more obscure In addition, studies with statisti- & Simes, 1997). The importance of
research. The exclusion of grey litera- cally significant findings are more funding suggests that the loss of valu-
ture raises questions about the simi- likely published in English (Tower of able grey literature may be especially
larity between grey literature and Babel bias) (Begg & Berlin, 1989; severe in nursing, where limited
more widely accessible research Egger & Smith, 1998). Researchers resources are available to fund
reports. are more likely to publish their statis- research. Unfortunately, none of the
tically significant findings in English studies examining correlates of publi-
journals (Egger et al., 1997). Research cation have focused on nursing sci-
Differences Between Grey reports with statistically significant ence. Documented differences
Literature and Widely findings are more likely to be pub- between grey literature and more eas-
lished in (a) journals with high cita- ily retrieved studies have fueled the
Disseminated Published
tion impact factors, (b) widely distrib- debate about whether grey literature
Literature uted journals, and (c) journals that are should be included.
If widely available literature were rep- indexed in computerized databases
resentative of all studies there would (Begg & Berlin, 1989; Egger & Smith,
be little need to include grey literature. 1998). Studies with statistically signif- Debate Regarding
Unfortunately, evidence suggests that icant findings are more likely pub- Inclusion of Grey
grey literature may differ in important lished repeatedly (duplicate reporting
Literature in
ways from research published in well- bias, Easterbrook, Berlin, Gopalan, &
known journals. The most consistent, Matthews, 1991). All of these factors
Meta-Analyses
noteworthy, and persistent difference could lead to different results from Some researchers have questioned
between published and unpublished studies that were retrieved and those whether it is proper to include in
research is that published research is that were not retrieved. Publication meta-analyses studies that have not
more likely to report findings that are bias is a term often used to refer to undergone peer review. The peer
258 Grey Literature in Meta-Analyses Nursing Research July/August 2003 Vol 52, No 4

review system is meant to assure that dence, which may introduce system- yield wider confidence intervals sur-
published articles meet widely atic error and thus threaten validity rounding estimates of effect size cen-
accepted methodological standards. (Moher et al., 2000). Since publica- tral tendency than if the grey literature
Hence, the central rationale for tion bias is most clearly and consis- were included. Researchers have
excluding grey literature is that tently linked with the significance of examined the impact of excluding
unpublished articles either (a) have findings, analyses that exclude unpub- grey literature on overall effect size
not passed peer review, and are of lished results risk overestimating estimates. However, the most useful
questionable scientific quality, or (b) effect sizes (McAuley et al., 2000). findings from meta-analysis often
have not undergone peer review at all, Lipsey and Wilson (1993) found that, come from moderator analysis (e.g.,
so their quality has not been assessed among 92 meta-analyses presenting whether effect sizes vary with inter-
(Sacks, Reitman, Pagano, & Kupel- separate effect size estimates for pub- vention characteristics). Small pilot
nick, 1996). Indeed, journals often lished and unpublished research, the studies with difficult-to-recruit sub-
exclude all but peer-reviewed cita- estimates from published sources were jects may be especially important in
tions. Essentially, the strategy of about one-third larger than those moving science forward but are diffi-
excluding unpublished literature from unpublished studies. Further, cult to access. The absence of such
treats publication status as a proxy for several studies have documented studies from meta-analysis limits the
study quality. Research suggests that changes in overall conclusions when usefulness of findings.
published and unpublished research grey literature reports were included
does not differ in scientific quality
(Chalmers et al., 1990; Easterbrook et Discussion of Strategies to
al., 1991). For instance, studies com-
Manage the Grey
paring published and unpublished
research find random assignment to
Literature Challenge
conditions unrelated to publication Increasing Access to Diverse Primary
(Dickersin & Min, 1993). Not all Studies
unpublished studies are of poor qual- Meta-analysis research Both individual scientists and dissemi-
ity, and not all published studies are of nation systems must address the issue
high quality. The quality of research, provides a quantitative of making diverse research findings
published or not, is distributed along review of existing accessible. Although publication bias
overlapping continua. Other critics of is a complex phenomenon, the prob-
grey literature have suggested that empirical evidence lem includes both researchers’ habits
meta-analyses should not give unpub- in submitting articles and editors’
lished literature the same weight as habits when evaluating them (Dick-
published studies (Sack et al., 1996). ersin & Min, 1993; Dickersin, 1997;
They suggest that meta-analysts con- Mahoney, 1977; Stern & Simes,
sider publication status when they cal- 1997). Cooper et al. (1997) found
culate average effect sizes. that the majority of studies with sig-
Proponents of the use of grey liter- nificant findings (74%) were submit-
ature claim that, ideally, meta-analysts as primary studies (Gregoire et al., ted for publication, but only a small
search out all studies that answer the 1995; Simes, 1987; Sutton, Duval, number of studies with nonsignificant
question because narrow searches Tweedie, Abrams, & Jones, 2000). findings (4%) were submitted. Sug-
may generate studies that are not rep- Differences between findings when gested reasons for investigators’
resentative of the research as a whole grey literature is included or excluded behavior include concern for one’s sci-
(Jadad, Moher, & Klassen, 1998). are not consistent with some studies entific reputation, lack of interest in
While many voice support for inclu- finding no differences (Copas & Shi, null results, and the belief that jour-
sion of grey literature, most published 2001; Fergusson, Laupacis, Salmi, nals will not accept such manuscripts
meta-analyses do not include grey lit- McAlister, & Huet, 2000; Juni, (Begg & Berlin, 1989; Boissel &
erature (McAuley et al., 2000). This Holenstein, Sterne, Bartlett, & Egger, Haugh, 1993). Editors and reviewers
exclusion may result from the difficul- 2002). Meta-analyses that have com- must make difficult decisions about
ties in obtaining grey literature stud- pared results of English only to lan- which studies to publish because jour-
ies. The value of grey literature must guage inclusive reports have also nal space is limited. Thornton and Lee
be high to justify the increased costs yielded inconsistent findings (Gre- (2000) suggested systematic publica-
of securing these difficult-to-locate goire et al., 1995; Juni et al., 2002; tion of abstracts of studies without
studies. Moher et al., 2000). It is possible that significant findings, which would
these differences reflect varied impact enable interested readers to contact
Consequences of Excluding of publication bias and language bias the investigators for further informa-
related to specific areas of science tion. Online journals with fewer space
Grey Literature From
(Juni et al., 2002). constraints could allow dissemination
Meta-Analyses Excluding grey literature also can of more studies without statistically
Excluding grey literature limits diminish the precision of meta-ana- significant results (Dickersin & Min,
reviews to a portion of available evi- lytic results. A restricted search may 1993).
Nursing Research July/August 2003 Vol 52, No 4 Grey Literature in Meta-Analyses 259

Investigators’ hesitancy to submit (g) examination of presentation practice of combining measures of


studies without significant findings is abstracts, (h) Internet searches, and (i) internal and external validity into a
likely to continue. Research registries contact with sources of synthesized single study quality score is question-
could render more studies easier to information. Detailed information able. Emerging work developing
find, in part because registration of about the sensitivity and specificity of scales with validated subscales to
trials occurs before studies are com- search strategies is beyond the scope address differing dimensions of qual-
pleted (Egger & Smith, 1998). The of this paper and addressed elsewhere ity may ameliorate problems with
Cochrane Collaboration has insti- (Conn et al., 2003). Once grey litera- existing scales. For example, the What
tuted registries for trials within its ture has been located, decisions Works Clearinghouse funded by the
areas of synthesis. In Australia, all should be made about inclusion and United States Department of Educa-
studies approved by ethics committees management of the primary research. tion is currently developing a set of
are enrolled in a prospective registry Concerns about the quality of pri- standards for evaluating the validity
(Stern & Simes, 1997). However, vol- mary studies often form the justifica- of causal claims. Their instrument to
untary registries cannot solve the tion for excluding grey literature. assess methodological quality will
problem because researchers may lack Indeed, a systematic approach to possess subscales such as intervention
sufficient incentive to participate addressing study quality across all construct validity, comparability of
(Thornton & Lee, 2000). Funding potential primary studies is consistent treatment groups, contamination,
agencies might demand full publica- with the scientific process. Explicit a outcome measure construct validity,
tion of findings (Dickersin & Min, priori inclusion criteria are essential. and statistical validity (www.w-w-
1993). The National Cancer Institute Methodological rigor may be c.org). These advances in measuring
and the cancer research community addressed through exclusion charac- study quality will enhance meta-ana-
might be the logical place to develop teristics directly related to study qual- lysts’ ability to use aspects of study
these strategies, given their experience ity (Lipsey & Wilson, 1993). Several quality for valid inclusion decisions,
with registries (Begg & Berlin, 1989). strategies used by meta-analysts to weight effect size estimates by
Other strategies may foster the address the fact that studies eligible methodological feature scores, or to
dissemination of more findings. For for inclusion vary in quality (Wort- examine methodological quality as a
example, reducing the emphasis on man, 1994). One strength of meta- moderator of effect size.
significance testing may help. Publica- analysis is the examination as an Despite extensive search strate-
tion bias possibly may be a reaction to empirical question of the influence of gies, it is impossible to locate all stud-
the dichotomous nature of hypothesis methodological quality or individual ies that address a particular research
testing that allows researchers to attributes of study design (e.g., the use question. Meta-analysts have pro-
interpret only significant findings of random assignment to conditions) posed numerous strategies to address
(Easterbrook et al., 1991). on effect size estimates. This allows difficulties resulting from problems in
Researchers, human subjects commit- conclusions about how design fea- obtaining a representative sample of
tees, funding agencies and editors tures are related to effect sizes to be existing studies. One long-standing
should accept responsibility for reduc- based on direct evidence, rather then a strategy is to calculate the fail-safe N,
ing the censoring of research results questionable proxy measure (i.e., pub- which identifies the number of hypo-
(Chalmers, 1990; Easterbrook et al., lication status). thetical no-effect studies required to
1991). Underreporting of research Unfortunately, there is very little bring a significant overall p level to
results is far more likely to have agreement in the research community nonsignificance (Orwin & Cordray,
adverse consequences for healthcare about measuring research quality. 1985). However, this strategy has
than publication of deliberately falsi- Many scales measuring study quality been replaced by more sophisticated
fied data (Chalmers, 1990). Thus, exist, but the authors of these scales techniques. Graphic presentation of
wider and more consistent dissemina- rarely provide a full demonstration of funnel plots has been used extensively
tion of all research is essential to their construct validity. In addition, to determine if a biased sample has
ensure that the scientific community even though study quality is probably been obtained. A funnel graph plots
takes advantage of the cumulative multidimensional, most quality scales the obtained effects sizes of the indi-
power of research for the benefit of either (a) ignore some dimensions of vidual trials against sample size.
humanity. study quality or (b) combine the dif- Soeken and Sripusanapan (2003) pro-
ferent dimensions of quality into a sin- vide an excellent overview of methods
Meta-Analytic Strategies to Address gle score. As an example of the latter, to detect publication bias. Generally,
Grey Literature Challenges a study quality scale may have ques- these strategies may suggest the pres-
Synthesists have long recognized that tions pertaining to the internal and ence of publication bias, but the
accessing diverse studies is a major external validity of studies. According nature of the bias remains unclear.
challenge. Valuable search strategies to some explicit or unstated algo- Since no entirely satisfactory sta-
include: (a) examination of multiple rithm, these answers then generate a tistical methods are available to
diverse computerized databases, (b) single score to represent the study’s address missing studies in meta-analy-
ancestry searches, (c) citation index quality. However, internal and exter- sis, inclusion of at least some grey lit-
searches, (d) examination of research nal validity are theoretically indepen- erature is essential to allow meta-ana-
registries, (e) journal hand searches, dent and in practice probably mod- lysts to compare effect sizes or other
(f) contact with the invisible college, estly and negatively correlated, so the study attributes to determine if they
260 Grey Literature in Meta-Analyses Nursing Research July/August 2003 Vol 52, No 4

systematically differ based on publica- for nursing impact. Nursing Outlook, of Internet searches to identify unpub-
tion or distribution status. Although 44, 267-271. lished clinical trials for systematic
some grey literature is difficult to Conn, V., Isamaralai, S., Rath, S., Jan- reviews. Medical Informatics & the
access, other sources of unpublished tarakupt, P., Wadhawan, R., & Dash, Internet in Medicine, 26, 203-218.
Y. (2003). Beyond MEDLINE for litera- Fergusson, D., Laupacis, A., Salmi, L.,
studies require few resources. For
ture searches. Journal of Nursing Schol- McAlister, F., & Huet, C. (2000). What
example, dissertations may be a rich arship, 35, 177-182. should be included in meta-analyses?
source of studies in some areas of sci- Conn, V., Valentine, J., & Cooper, H. An exploration of methodological
ence. A recent meta-analysis included (2002). Interventions to increase physi- issues using the ISPOT meta-analyses.
10 dissertations in a project with 43 cal activity among aging adults: A meta- International Journal of Technology
primary studies (Conn et al., 2002). analysis. Annals of Behavioral Medi- Assessment in Health Care, 16, 1109-
Obtaining major funding for meta- cine, 24, 190-200. 1119.
analysis research will allow searches Cook, A. M., Finlay, I. G., Edwards, A. G. Gregoire, G., Derderian, F., & Le Lorier, J.
to retrieve grey literature. K., Hood, K., Higginson, I. J., Good- (1995). Selecting the language of the
Meta-analyses are increasingly win, D. M., et al. (2001). Efficiency of publications included in a meta-analy-
searching the grey literature in palliative sis: Is there a Tower of Babel bias? Jour-
forming the scientific foundation for
care. Journal of Pain & Symptom Man- nal of Clinical Epidemiology, 48, 159-
nursing practice and helping to set its agement, 22, 797-801. 163.
future research agenda. Results Cooper, H., DeNeve, K., & Charlton, K. Hubbard, R., & Armstrong, J. S. (1997).
obtained with this method are more (1997). Finding the missing science: Publication bias against null results.
precise than narrative reviews. How- The fate of studies submitted for Psychological Reports, 80, 337-338.
ever, results of meta-analyses must be review by a human subjects committee. Jadad, A. R., Moher, D., & Klassen, T. P.
interpreted within the context of the Psychological Methods, 2, 447-452. (1998). Guides for reading and inter-
limitations of their primary studies. Cooper, H., & Hedges, L. (Eds.) (1994). preting systematic reviews: II. How did
Publication bias, realized though the The handbook of research synthesis. the authors find the studies and assess
exclusion of grey literature, can be a New York: Russell Sage Foundation. their quality? Archives of Pediatrics &
Copas, J., & Shi, J. (2001). A sensitivity Adolescent Medicine, 152, 812-817.
powerful limitation on the validity
analysis for publication bias in system- Juni, P., Holenstein, F., Sterne, J., Bartlett,
and value of meta-analytic results. atic reviews. Statistical Methods in C., & Egger, M. (2002). Direction and
The goal of meta-analysis—to system- Medical Research, 10, 251-265. impact of language bias in meta-analy-
atically review all evidence—requires Dickersin, K. (1997). How important is ses of controlled trials. International
that as many primary studies as possi- publication bias? A synthesis of avail- Journal of Epidemiology, 31, 115-123.
ble be identified and included in the able data. AIDS Education & Preven- Lipsey, M. W., & Wilson, D. B. (1993).
quantitative synthesis. ▼ tion, 9(Supp Al), 15-21. The efficacy of psychological, educa-
Dickersin, K. (1994). Research registers. In tional, and behavioral treatment. Con-
Accepted for publication March 24, 2003. H. Cooper & L. Hedges (Eds.), The firmation from meta-analysis. American
handbook of research synthesis (pp. 71- Psychologist, 48, 1181-1209.
Financial support provided by a grant 83). New York: Russell Sage Founda- Mahoney, M.J. (1977). An experimental
from the NIH NINR (RO1NR07870) to
Vicki Conn, principal investigator. tion. study of confirmatory bias in the peer
Dickersin, K., Chan, S., Chalmers, T. C., review system. Cognitive Therapy and
Corresponding author: Vicki S. Conn, Sacks, H. S., & Smith, H., Jr. (1987). Research, 1, 161-175.
PhD, RN, S317 School of Nursing-MU,
Columbia, MO 65211 (e-mail: conn@ Publication bias and clinical trials. Con- McAuley, L., Pham, B., Tugwell, P., &
missouri.edu) trolled Clinical Trials, 8, 343-353. Moher, D. (2000). Does the inclusion of
Dickersin, K., & Min, Y. I. (1993). NIH grey literature influence estimates of
clinical trials and publication bias. intervention effectiveness reported in
Online Journal of Current Clinical Tri- meta-analyses? Lancet, 356 (9237),
References als, Doc. No. 50. 1228-1231.
Begg, C. B., & Berlin, J. A. (1989). Publi- Dickersin, K., Min, Y. I., & Meinert, C. L. Moher, D., Cook, D. J., Eastwood, S.,
cation bias and dissemination of clinical (1992). Factors influencing publication Olkin, I., Rennie, D., & Stroup, D. F.
research. Journal of the National Can- of research results. Follow-up of appli- (2000). Improving the quality of reports
cer Institute, 81, 107-115. cations submitted to two institutional of meta-analyses of randomised con-
Boissel, J. P., & Haugh, M. C. (1993). The review boards. Journal of the American trolled trials: The QUOROM state-
iceberg phenomenon and publication Medical Association, 267, 374-378. ment. QUOROM Group. British Jour-
bias: The editors’ fault? Clinical Trials Easterbrook, P. J., Berlin, J. A., Gopalan, nal of Surgery, 87, 1448-1454.
& Meta-Analysis, 28, 309-315. R., & Matthews, D. R. (1991). Publica- Orwin, R. G., & Cordray, D. S. (1985).
Chalmers, I. (1990). Underreporting tion bias in clinical research. Lancet, Effects of deficient reporting on meta-
research is scientific misconduct. Jour- 337, 867-872. analysis: a conceptual framework and
nal of the American Medical Associa- Egger, M., & Smith, G. D. (1998). Bias in reanalysis. Psychological Bulletin, 97,
tion, 263, 1405-1408. location and selection of studies. British 134-147.
Chalmers, I., Adams, M., Dickersin, K., Medical Journal, 316, 61-66. Sacks, H. S., Reitman, D., Pagano, D., &
Hetherington, J., Tarnow-Mordi, W., Egger, M., Zellweger-Zahner, T., Schneider, Kupelnick, B. (1996). Meta-analysis:
Meinert, C., et al. (1990). A cohort M., Junker, C., Lengeler, C., & Antes, G. An update. Mount Sinai Journal of
study of summary reports of controlled (1997). Language bias in randomised Medicine, 63, 216-224.
trials. Journal of the American Medical controlled trials published in English and Simes, R. J. (1987). Confronting publica-
Association, 263, 1401-1405. German. Lancet, 350, 326-329. tion bias: A cohort design for meta-
Conn, V. S., & Armer, J. M. (1996). Meta- Eysenbach, G., Tuische, J., & Diepgen, T. analysis. Statistics in Medicine, 6, 11-
analysis and public policy: Opportunity L. (2001). Evaluation of the usefulness 29.
Nursing Research July/August 2003 Vol 52, No 4 Grey Literature in Meta-Analyses 261

Soeken, K., & Sripusanapan, A. (2003). cation in a cohort study of clinical Statistical Methods in Medical
Assessing publication bias in meta- research projects. British Medical Jour- Research, 9, 421-445.
analysis. Nursing Research, 52, 57-60. nal, 315, 640-645. Thornton, A., & Lee, P. (2000). Publica-
Sterling, T., Rosenbaum, W., & Weinkam, Sutton, A. J., Duval, S. J., Tweedie, R. L., tion bias in meta-analysis: Its causes
J. (1995). Publication decisions revis- Abrams, K. R., & Jones, D. R. (2000). and consequences. Journal of Clinical
ited: The effect of the outcome of statis- Empirical assessment of effect of publi- Epidemiology, 53, 207-216.
tical tests on the decision to publish and cation bias on meta-analyses. British Wortman, P. M. (1994). Judging research
vice versa. American Statistician, 49, Medical Journal, 320, 1574-1577. quality. In H. Cooper & L. V. Hedges
108-112. Sutton, A. J., Song, F., Gilbody, S. M., & (Eds.), The handbook of research syn-
Stern, J. M., & Simes, R. J. (1997). Publi- Abrams, K. R. (2000). Modelling publi- thesis (pp. 97-109). New York: Russell
cation bias: Evidence of delayed publi- cation bias in meta-analysis: A review. Sage Foundation.

You might also like