European Accounting Review Vol. 17, No. 2, 305 –330, 2008
Meta-analysis and the Accounting Literature: The Case of Audit Committee Independence and Financial Reporting Quality BRADLEY POMEROY and DANIEL B. THORNTON
School of Business, University of Alberta, Edmonton, Canada and School of Business, Queen’s University, Kingston, Canada
ABSTRACT We conduct a meta-analysis (MA) of the association between audit committee (AC) independence and financial reporting quality (FRQ). Although we cannot reliably aggregate results across studies in a statistical sense because of inconsistencies in defining FRQ and the absence of replication studies, quantitative review techniques yield three conclusions: (1) The use of different FRQ measures in the AC independence literature explains about half of the variation in results across studies. (2) Audit committees are more effective at enhancing audit quality (e.g. through averting going-concern reports and auditor resignations) than they are at fostering financial statement quality (e.g. by making high quality accruals and avoiding restatements). AC independence can even reduce apparent financial statement quality by identifying the need for restatements and remedial, abnormal accruals. (3) Financial statement quality and audit quality are complementary contributors to FRQ. The statistical and methodological difficulties we encounter lead us to posit that the dearth of MA studies in accounting and auditing stems from similar difficulties in applying MA to other topics. We present evidence consistent with publication biases and perverse researcher incentives being responsible for the difficulties.
1.
Introduction
Although financial reporting quality is not directly observable, prominent commentators vigorously assert its importance as a capital market mainstay. For example, Arthur Levitt, former Chair of the US Securities and Exchange
Correspondence Address: Daniel B. Thornton, School of Business, Queen’s University, Kingston, Canada, K7L 3N6. Tel.: þ1 613 549 8212. E-mail:
[email protected] 0963-8180 Print/1468-4497 Online/08/020305–26 # 2008 European Accounting Association DOI: 10.1080/09638180701819832 Published by Routledge Journals, Taylor & Francis Ltd on behalf of the EAA.
306
B. Pomeroy & D. B. Thornton
Commission (SEC), says ‘high quality accounting standards . . . improve liquidity [and] reduce capital costs’ (Levitt, 1998) and ‘quality information is the lifeblood of strong, vibrant markets. Without it, liquidity dries up. Fair and efficient markets cease to exist’ (Levitt, 2000). The objectives of this study are twofold. First, we aim to apply meta-analysis, a quantitative review method, to summarize existing research findings examining the association between financial reporting quality and audit committee independence. We find that we can draw only limited conclusions about this association because of various obstacles we encounter in attempting to implement a full-blown meta-analysis. Second, we probe the likely sources of the obstacles encountered, concluding that today, accounting/ auditing publication outlets and researcher incentives are generally biased against introducing new evidence on old topics that corroborates or conflicts with prior findings. 1.1. Outline of the Study and Overview of Results We apply meta-analysis (henceforth, MA) to summarize findings relating to the association between financial reporting quality (FRQ) and audit committee (AC) independence, employing MA procedures outlined by Greenberg (1992) and Lipsey and Wilson (2001). This exercise is innovative, given the dearth of MA studies in accounting. It is apparently feasible, given DeZoort et al.’s observation (2002, p. 45) that AC member independence is ‘. . . [o]ne of the most common variables in the AC composition literature’. It is also potentially useful to managers, auditors, regulators and investors, given the scattered and often conflicting conclusions in the AC/FRQ literature to date. Managers and auditors can use reliable MA results to assess the financial reporting costs and benefits that are likely to stem from expenditures on AC independence. Regulators such as securities commissions can use reliable MA results to enhance confidence in identifying firms whose FRQ warrants scrutiny, directing scarce resources toward the most serious likely breaches of financial reporting standards. Investors can use reliable MA results to assess the risk inherent in the accounting information they rely on to make buy, sell and hold decisions regarding corporate securities. In the process of applying MA, we find that we cannot reliably aggregate across studies in a statistical sense because of significant inconsistencies in defining FRQ and the absence of studies designed to replicate, and thereby enhance confidence in, the results of previous studies. However, quantitative review techniques yield the following preliminary conclusions and issues for future research to explore: 1. The use of different FRQ measures in the AC independence literature is responsible for more than half the variation in results across studies. 2. Audit committees are more effective at enhancing audit quality (e.g. through averting going-concern reports and auditor resignations) than they are at fostering financial statement quality (e.g. by making high quality accruals and avoiding restatements). AC independence can even reduce apparent financial
Meta-analysis and the Accounting Literature
307
statement quality by identifying the need for restatements and remedial, abnormal accruals. 3. Financial statement quality and audit quality contribute to FRQ in different, complementary ways. Additional theory is needed to understand FRQ as a conceptual construct. We maintain that MA’s ability to yield such preliminary conclusions and to highlight new research issues is a significant by-product of MA that can help to advance any literature, including the FRQ/AC literature, even when reliable statistical aggregation across studies is not feasible. The inability to perform a statistically reliable MA in such a straightforward venue leads us to posit that the dearth of MA studies in the accounting and auditing literatures compared with other business and non-business disciplines stems from similar difficulties in applying MA to other accounting and auditing topics. We present evidence suggesting that publication biases and perverse researcher incentives, which are particularly acute in the accounting/auditing literatures, are consistent with such difficulties. 1.2. Organization of the Study Section 2 outlines MA’s contributions to accounting/auditing and other disciplines to date. The relative dearth of MA studies in the accounting literature suggests the existence of either unexploited opportunities or significant barriers to applying the technique. In Section 3, a meta-analysis of the association between AC independence and FRQ, with mathematical details of the exercise relegated to an appendix, summarizes existing studies and highlights the difficulties we encountered in applying MA to aggregate the often conflicting results of extant studies. Section 4 presents evidence consistent with the existence of publication biases and disincentives to conducting MA studies or replication studies that would be amenable to MA when combined with seminal contributions to the literature. Section 5 offers conclusions and suggestions for future research. 2.
Prior Literature
2.1. Benefits of Meta-analysis Meta-analysis is a quantitative review method for standardizing and aggregating findings across empirical studies (Greenberg, 1992). Lipsey and Wilson (2001, p. 2) tout MA as an effective way to ‘. . . summarize, integrate, and interpret selected sets of scholarly works in the various disciplines’. Smith and Glass (1977) introduced MA in a seminal clinical psychology study to determine whether psychotherapy improves patients’ conditions on average. The study aggregated and standardized the results of 375 extant papers reporting positive/negative and significant/non-significant results to determine a mean
308
B. Pomeroy & D. B. Thornton
treatment effect. Following this study, researchers debated the merits of MA. Some argue that aggregating the results of studies varying in quality leads to unreliable generalizations – a garbage-in-garbage-out effect. However, MA is widely accepted and used in medical research and disciplines other than accounting and auditing. Given the disparate results contained in experimental studies of treatments for socially critical medical illnesses like cancer and heart disease, MA is particularly helpful to medical professionals who canvass research results in deciding how to treat patients (e.g. LeLorier et al., 1997). MA facilitates this reliance by providing systematic, quantitative literature reviews that enhance the precision of effect size estimates and hence the reliability of research findings. In principle, MA can reliably aggregate the data contained in an unlimited number of related studies, thereby increasing sample sizes and statistical power and elucidating mean relationships among key variables. MA is particularly effective in reconciling results that are inconsistent across studies and educing significant relationships among variables in a group of studies whose individual sample sizes are too small to support reliable inferences. In both instances, MA can effectively combine the samples and determine reliable, overall relationships among key variables. Thus, we expect to observe extensive use of MA in the accounting and auditing literatures to document reliable effects of various treatments, such as the effect of balanced-score-card reporting on employee motivation and the effect of audit committee independence on FRQ. In practice, however, we find that the use of MA and quantitative reviews in general are uncommon in the accounting and auditing literatures. 2.2. Meta-analyses in Accounting and Auditing Literatures Greenberg (1992) provides instructions for conducting MA, outlines its advantages over normative techniques in generating rigorous and unbiased literature reviews that would clearly identify opportunities to extend the relevant literature, and encourages accounting and auditing Ph.D. students to use MA in conducting literature reviews and developing thesis proposals. These advantages include generating a combined significance test and average effect size to determine whether a literature in aggregate has rejected a null hypothesis, examining variability between studies within a literature to identify potential moderating variables to include in future studies, and determining sample sizes needed to provide sufficient power in future studies. Despite its alleged advantages, accounting researchers have seldom exploited MA in prior research. In August 2007 we searched ABI Inform (Keywords: Metaanalysis AND Accounting OR Auditing) for prior accounting studies that employed MA, garnering 33 hits, 14 of which are MA studies closely related to accounting phenomena and 12 of which are published in accounting or auditing journals.1 To compare the incidence of MA studies in accounting to that in other business disciplines, we searched for Meta-analysis AND Marketing
Meta-analysis and the Accounting Literature
309
(105 hits); Meta-analysis AND Finance (18 hits); Meta-analysis AND Management (233 hits). These results are consistent with the primary targets for MA research being non-archival, experimental and quasi-experimental behavioral studies, whose research designs are similar to those that spawned MA in the psychology and medical literatures. However, nearly all MA studies published in accounting journals, and many MA studies published in other business disciplines, aggregate the results of empirical/archival literatures. Thus, the empirical/archival nature of accounting studies, per se, cannot account for the relative paucity of MA in accounting and auditing. Of the accounting-related MA studies, only three appear in the top-tier accounting and auditing journals. Trotman and Wood (1991) conduct a MA of 14 published and 3 unpublished studies within the internal control judgment literature. Hay et al. (2006) perform a MA on 147 empirical analyses conducted over 25 years in the audit-fee literature. Kinney and Martin (1994) conduct a MA of the impact of auditing on accounting adjustments, aggregating the results of nine archival data sets and more than 1,500 audits. Other MA studies published in accounting journals assess the following literatures: accountants’ job satisfaction (Brierley, 1999; Dole and Schroeder, 2001), the association between corporate characteristics and disclosure levels (Ahmed and Courtis, 1999), multinational transfer pricing (Borkowski, 1996), selectivity performance of UK unit trusts (Fletcher, 1995), the Foreign Corrupt Practices Act (Cooper et al., 1985), budgeting in prison management (Zager et al., 2001) and participative budgeting (Greenberg et al., 1994).2 Thornton (1994) employs MA to combine the results of several papers comprising a Canadian Academic Accounting Association education project. Ostensibly belying its objective of serving as an outlet for rigorous review articles, The Journal of Accounting Literature has published only one MA study since its inception in 1982 (Greenberg et al., 1994). The benefits of MA and its scarcity in the accounting/auditing literature, together with the alleged social importance of FRQ and the consequent need to understand its nature and its determinants, motivate us to attempt a meta-analysis of the relation between AC independence and FRQ.
3.
Meta-analysis of the Relation between Audit Committee Independence and Financial Reporting Quality
3.1. Identification of Prior Studies Table 1 summarizes the results of our literature search, which gleaned 27 published and unpublished (working) papers examining the association between FRQ and AC independence. Twenty-seven studies should provide ample scope for conducting MA. To assemble the entries in Table 1, we began by searching ABI Inform and the Social Sciences Research Network (SSRN) for AC independence studies
Author
Published?
Sample size
Dependent variable
Independent variable
1
Klein
Yes (2002) 692 firm-years
Abnormal accruals
2
Xie et al.
Yes (2003) 282 firm-years
Discretionary current accruals % INDAC
3
Be´dard et al.
4
Williams
5
Agrawal and Chadha Abbott et al.
Yes (2004) 200 aggressive and 100 control firms No (2002) 465 firm-years (Singapore) Yes (2005) 159 matched pairs
Income-increasing abnormal 100% INDAC accruals Abnormal accruals AC familiarity with auditor Earnings restatements % INDAC
Yes (2004) 88 matched pairs
Restatement firms
Yes (2005) 896 firm-years
Quarterly total discretionary % INDAC accruals (AAERs) 100% INDAC
6 7 8
Yang and Krishnan Abbott et al.
Yes (2000) 78 matched pairs
9
Lee et al.
Yes (2004) 190 matched pairs
10
Farber
Yes (2005) 39 matched pairs
11
Yes (2004) 1,052 firm-years
12
Anderson et al. Bryan et al.
No (2004) 1,291 firm-years
13
Lennox
No (2002) 457 firms
14
Anderson et al.
No (2003) 1,241 firms
Number of auditor-initiated resignations (AAERs)
% INDAC
100% INDAC
100% INDAC
No. of outside directors on AC Cost of debt financing (yield 100% INDAC spread) Unexpected earnings 100% INDAC
No. INDAC members leaving No. INDAC after auditor dismissal members before dismissal Cumulative abnormal return % INDAC
Statistic
Conclusion
Chi-square Negative significant t-stat. Negative not significant p-value Negative significant t-stat. Negative significant p-value Negative not significant Chi-square Negative significant t-stat. Positive not significant Chi-square Negative significant t-stat. Negative significant t-stat. Positive not significant t-stat. Negative significant t-stat. Positive significant Z-stat. Negative significant p-value
Positive significant
B. Pomeroy & D. B. Thornton
Number
310
Table 1. Papers included in the meta-analysis
15
Felo et al.
No (2003) 119 firms
16
Wright
No (1996)
17
Bradbury et al. No (2004)
18
Carcello and Yes (2003) Neal Raghunandan Yes (2003) and Rama
19 20
Abbott et al.
21
Carcello et al. Yes (2002) 254 firms
22 23
Carcello and Neal Persons
Yes (2000) 223 financially distressed Going-concern-modified firms reports Yes (2005) 111 matched pairs (AAERs)
% affiliated AC members 100% INDAC
24
Parker et al.
25
Ebrahim
Yes (2005) 161 financially distressed Going-concern-modified firms reports Yes (2007) 2,360 firm-years Abnormal accruals
% affiliated AC members 100% INDAC
26
Lin et al.
Yes (2006) 106 matched pairs
100% INDAC
27
Rahman and Ali
Yes (2006) 194 firm-years (Malaysia) Abnormal accruals
Audit fees
Earnings restatements
% INDAC
% INDAC
t-stat.
Negative not significant t-stat. Positive significant p-value Negative significant Chi-square Positive significant p-value Positive significant t-stat.
Positive significant p-value Positive significant Chi-square Negative significant Z-stat. Negative significant Z-stat. Negative significant t-stat. Negative not significant Chi-square Positive not significant p-value Negative not significant
Notes: % INDAC ¼ percentage of independent members on the AC; 100% INDAC ¼ fully independent AC; AAER ¼ SEC Accounting and Auditing Enforcement Releases; Published? ¼ ‘no’ denotes an unpublished working paper.
Meta-analysis and the Accounting Literature
Yes (2003)
Perceived financial reporting % INDAC quality 34 matched pairs (AAERs) % insider/grey AC members 252 firms (Singapore and Income-increasing accruals 100% INDAC Malaysia) 374 firms Auditor dismissals following % affiliated AC going-concern report members 124 firms % votes against auditor 100% INDAC with ratification high non-audit fee ratio 492 firms Audit fees 100% INDAC
311
312
B. Pomeroy & D. B. Thornton
(Keywords: Audit committee AND Independence). To obtain additional studies, we also searched conference databases and downloaded researcher CVs. This search yielded 27 papers on topics closely related to the association between AC independence and FRQ, 20 published and seven unpublished. Table 1 summarizes the studies and their conclusions. Canvassing unpublished (working) papers can lead to both strengths and weaknesses in MA. In principle, aggregating both published and unpublished results adumbrates potential implications of statistically non-significant results, mitigating publication and replication bias. Unpublished studies, however, are likely to exhibit inconsistent research quality since they have not fully survived peerreview processes. We find that one of the unpublished papers reports statistically non-significant results and seven of the published papers report non-significant associations between AC independence and FRQ.3 3.2. Overview of the Literature Studies examining AC impact on FRQ generally posit that AC member characteristics, such as independence and financial expertise, are positively associated with FRQ. In a conventional (i.e. non-MA) literature review, DeZoort et al. (2002, p. 38) state, ‘[t]he empirical audit committee literature is both diverse and expansive, with rapid growth in recent years based on increased concerns about corporate governance and the quality of financial reporting’. They add that ‘[o]ne of the most common variables in the audit committee composition literature is member independence’ (DeZoort et al., 2002, p. 45). Thus, given the frequency of studies and the importance of the topic to securities regulators, the AC independence literature appears to be a viable and worthy MA candidate. The AC literature evinces a notable lack of consensus as to the proper variables to use as proxies for FRQ and its determinants. Klein (2002) reports that AC independence is negatively associated with the level of discretionary accruals, whose unrestricted use is thought to reduce FRQ (Dechow and Dichev, 2002). She also finds that, whereas regulators favor a fully independent AC, an AC composed of only a majority of independent directors is more likely to curb discretionary accruals than a fully independent committee. In contrast to Klein (2002), Farber (2005) does not find any significant association between the proportion of independent AC members and the incidence of SEC accounting and auditing enforcement releases, which presumably stem from poor FRQ. However, Farber’s proxy for poor FRQ differs from Klein’s (2002), and is likely to derive from more severe forms of earnings manipulation than the abuse of discretionary accruals, such as the recognition of bogus revenues from sales to related parties. Using yet another proxy for FRQ, Abbott et al. (2004) report that firms with fully independent ACs are much less likely to restate earnings than those without fully independent ACs. Although earnings restatements are likely to stem from low FRQ, companies often restate earnings without any fillip from
Meta-analysis and the Accounting Literature
313
market regulators like the SEC. Thus, to the extent that the SEC directs its scarce resources to remedying the most visible and extreme financial reporting abuses, this third proxy for FRQ differs from the previous two. Be´dard et al. (2004, p. 35) conclude that AC independence ‘. . . contribute[s] significantly to effective control of aggressive earnings management’. Williams (2002) and Bradbury et al. (2004) find similar, significant relationships in samples of 465 Singapore firm-years and 252 Singapore/Malaysia firms, respectively. In contrast, Xie et al. (2003) do not find any significant relationship between the proportion of independent AC members and the incidence of aggressive earnings management in a sample comprised of 280 US firm-years. We conclude that the AC independence literature exhibits strikingly inconsistent conclusions. This inconsistency stems at least partly from the diverse FRQ proxies used in the various studies. Indeed, we are unable to identify a dominant or consistent measure for this construct, ‘financial reporting quality’, in the literature. 3.3. Categorization of Proxies for Financial Reporting Quality Figure 1 summarizes the variables used in extant literature as proxies for FRQ. Initially we attempted to rank-order the proxies in terms of the degree to which they indicate low FRQ. Although we could not agree on an ordinal ranking, we concluded that we could allocate FRQ proxies to high, low and very low categories as follows: Proxies for high quality financial reporting: 1. Perceived financial reporting quality (Felo et al., 2003) 2. Cumulative abnormal returns after controlling for other factors (Anderson et al., 2003; Bryan et al., 2004) 3. Low cost of debt financing (Anderson et al., 2004). Proxies for low quality financial reporting: 1. Number of independent AC members that leave after an auditor dismissal (Lennox, 2002) 2. Audit fees after controlling for firm size and other variables, the idea being that more independent ACs insist on higher quality, costlier audits (Carcello et al., 2002; Abbott et al., 2003) 3. Proportion of votes against auditor ratification (Raghunandan and Rama, 2003) 4. Presence of going-concern reports (Carcello and Neal, 2000; Parker et al., 2005) 5. Earnings management measured by abnormal accruals (Klein, 2002; Williams, 2002; Yang and Krishnan, 2005; Rahman and Ali, 2006; Ebrahim, 2007)
314 B. Pomeroy & D. B. Thornton Figure 1. Tentative ranking of financial reporting quality proxies. Notes: This is a three-level ranking; hence, we do not differentiate financial reporting proxies within each level. The numbers link the financial reporting quality measures with their corresponding empirical studies. AAER ¼ SEC Accounting and Auditing Enforcement Release.
Meta-analysis and the Accounting Literature
315
6. Aggressive earnings management (Xie et al., 2003; Be´dard et al., 2004; Bradbury et al., 2004) 7. Auditor resignations or dismissals (Carcello and Neal, 2003; Lee et al., 2004). Proxies for very low quality financial reporting: 1. Presence of earnings restatements (Abbott et al., 2004; Agrawal and Chadha, 2005; Lin et al., 2006) 2. SEC issuance of accounting and auditing enforcement releases (AAERs) (Wright, 1996; Abbott et al., 2000) 3. SEC issuance of AAERs relating to fraudulent reporting (Farber, 2005; Persons, 2005). Each study rationalizes the dependent variable chosen but these diverse measures obviously differ in the level of FRQ they proxy for. Thus, one might expect to see some effort to reach consensus in the literature. Published replication studies, however, are virtually non-existent in this literature and the most commonly employed measure of FRQ, earnings management (using abnormal accruals as the proxy), has attracted criticism.4 3.4. Meta-analysis Procedures and Results Despite the data limitations, we perform MA on the association between AC independence and FRQ as if the FRQ measures were comparable across studies, relegating the mathematical details of the exercise to the Appendix. Using the standard quantitative techniques of MA, we empirically test and reject the hypothesis that these seemingly diverse measures capture the same FRQ construct. In addition, we examine whether the literature as a whole has rejected the null hypothesis that AC independence has no significant impact on FRQ. Briefly, the steps involved (outlined in detail by Lipsey and Wilson, 2001) are as follows: 1. Convert test statistics in the various studies, such as p-values and F-scores, to a standard correlation measure called ‘effect size’. 2. Compute mean effect sizes using weighting factors based on assumptions regarding the homogeneity of variances across studies. 3. Test the homogeneity hypothesis in 2 above. The results in the Appendix are mechanically generated, assuming the MA requirements are fulfilled. Of course, the inconsistent use of proxies just discussed suggests that the results in Step 2 are not reliable if Step 3 indicates a violation of homogeneity of variances (Greenberg, 1992; Lipsey and Wilson, 2001).
316
B. Pomeroy & D. B. Thornton
3.5. Heterogeneity of Proxies for Financial Reporting Quality Homogeneity-of-variances tests indicate that the FRQ proxies vary significantly ( p , 0.001). To further explore this result, we conduct an analysis of betweenstudy variability to identify any significant moderating variables as follows:5 1. Group the financial reporting proxies into common categories based on our ranking in Figure 1. 2. Calculate an overall effect size for each FRQ group (e.g. add together the effect sizes for all the studies that use accruals as the proxy for FRQ). 3. Calculate a Q-statistic for each group and test the homogeneity of variances assumption for each group. This analysis yields a total of six groups that do not violate the homogeneity of variances assumption, ranging from the highest quality proxies such as cumulative abnormal returns to the lowest quality proxies such as earnings restatements. 4. Sum the Q-values across all individual groups to obtain a within-group Q; then subtract the within-group Q from the overall Q-value obtained by the homogeneity of variances test to obtain a between-group Q measure. 5. Observe whether the between-group Q (i.e. the variability due to differences in FRQ proxies across the literature) accounts for a significant proportion of the overall variability across studies. This analysis of the between-study variability technique is similar to a one-way ANOVA. The results indicate that assigning FRQ proxies to six groups accounts for a significant proportion of the variability in the overall effect size ( p , 0.005). The between-group Q accounts for more than half (namely, 52.29%) of the variability in the overall effect size. This suggests that the various proxies for financial reporting quality used in the audit committee independence literature are likely tapping into different concepts, and this factor limits our ability to aggregate across these studies. This finding suggests that additional theorizing is needed to understand the concept of FRQ. Although a standard (non-MA) literature review could discuss the potential limitations of aggregating studies with different dependent variable proxies, it would not be capable of identifying, explaining and quantifying such aggregation problems.
3.6. Differences in the Association between Financial Reporting Quality Proxies and Audit Quality In a second MA application, we analyze the weighted effect size of each FRQ proxy group to see if there are any quantitative or qualitative differences in the association between AC independence and FRQ across the groups. The analysis, outlined in Table 2, indicates that the mean effect size is significant for all groups except the earnings restatement group. Comparing the mean effect size across groups, we find that the auditor ratification/going concern group has the strongest association,
Meta-analysis and the Accounting Literature
317
Table 2. Mean effect sizes and confidence intervals for financial reporting proxy groups Auditor ratification/ Auditor Earnings High quality going-concern reports Accruals resignations restatements AAERs proxiesa Mean ES SE U.L. L.L. Z-Score
0.090 0.014 0.118 0.062 6.311
0.203 0.054 0.310 0.097 3.756
0.048 0.019 0.084 0.011 2.556
0.181 0.037 0.252 0.109 4.942
0.048 0.045 0.137 20.041 1.062
0.176 0.059 0.292 0.061 2.988
Notes: Mean ES ¼ mean effect size; SE ¼ standard error; U.L. ¼ upper confidence limit; L.L. ¼ lower confidence limit. a The high quality proxies group includes the following financial reporting quality proxies: perceived financial reporting quality; cumulative abnormal returns; and cost of debt financing.
and the accruals and earnings restatement groups have the weakest association, with AC independence. Moreover, we generally find that the proxy groups relating to audit quality (i.e. auditor ratification, going concern, resignations and AAERs) have a greater effect size compared to those proxies that appear to relate more directly to FRQ as manifest in financial statements (i.e. high quality proxies, accruals and earnings restatements). If this result stood up to further scrutiny, it would suggest that independent AC members are more effective at improving audit quality than they are at enhancing financial statement quality. The results also suggest that an FRQ proxy is not necessarily a relevant response variable to use in studying audit quality. This observation, if corroborated by additional research, would imply that, cross-sectionally, increases in audit quality are not necessarily expected to be associated with increases in observed FRQ. Indeed, all else being equal, greater AC independence could enhance the likelihood of uncovering the need for restatements and remedial (abnormal) accruals that reduce observed FRQ as it is often characterized. Although other factors, such as differences in sample size, can contribute to such an observation, our MA and additional analysis using formal, quantitative review techniques yield several conjectures for future research to explore.
1. An archival study could develop a structural equations model to examine the direction of causality for audit quality and FRQ. For example, does stronger financial reporting enhance audit quality or does an effective audit team breed better financial reporting? 2. Archival studies could use different audit-strength variables and FRQ proxies to determine which proxies exhibit stronger associations and whether the FRQ proxies exhibit directional causality. For example, does aggressive earnings management using abnormal accruals result in more earnings restatements?
318
B. Pomeroy & D. B. Thornton
Do earnings restatements presage later accounting frauds; or do accounting frauds lead to earnings restatements? 3. An experimental study could determine whether audit committee members respond differently to general audit issues (e.g. ensuring that auditors can perform their duties without encountering management-initiated roadblocks) compared to financial reporting concerns (e.g. taking the auditors’ positions on complex accounting-policy decisions (Gibbins et al., 2007)).
3.7. Implications of the Meta-analysis Exercise In summary, we reject the hypothesis of homogeneity in the variances of FRQ proxies. Hence, even if the FRQ proxies were intuitively consistent, generalizing across studies in a strict statistical sense could not be supported by performing MA using extant data. However, our application of MA’s quantitative review techniques yields preliminary conclusions by formally identifying groups of FRQ proxies that are likely to measure similar constructs and documenting which constructs are likely to exhibit stronger associations with AC independence. Thus, we agree with Greenberg’s (1992) contention that MA can be useful to researchers by identifying important, moderating independent variables for future research to explore. We also maintain, however, that MA’s quantitative review techniques can assist researchers in understanding the impact of inconsistent dependent variable usage across the literature in at least two ways: (1) by quantifying the degree of heterogeneity in dependent variable proxies across the literature; and (2) by identifying differences in the degree to which the various proxies are associated with the explanatory variables we are interested in, particularly AC independence. We conclude that accounting/auditing research can benefit from applying MA techniques even when limitations inherent in the literature initially impede the process and preclude definitive generalizations that would stem from a full-blown, conventional MA.
4.
Barriers to Meta-analysis in Accounting and Auditing
Because of the difficulties encountered, we decided to explore potential reasons for these difficulties and for the dearth of MA studies in the accounting and auditing literatures generally. We see at least three candidate explanations for the diversity in FRQ measures: (1) researchers cannot agree on a single measure; (2) methodological limitations and construct validity concerns in all existing measures spawn demand for new measures; (3) publication biases and researcher incentives in the accounting research environment foment diversity and inconsistency in dependent variable selection. That is, novel studies, using original measures, are more likely to survive peer review than replication studies and studies using the same variables as previous studies. The confluence of evidence is consistent with the third explanation.
Meta-analysis and the Accounting Literature
319
4.1. Publication Bias Publication bias is ‘. . . the phenomenon in which studies with positive results are more likely to be published than studies with negative [inconclusive] results’ (Peterson, 1989, p. 5). Hence, papers published in major journals are unlikely to be representative of the population of extant empirical results obtained by researchers. This publication bias severely limits the potential contributions of MA studies focusing only on articles appearing in those journals. Some statistically nonsignificant, unpublished results are retrievable electronically or by corresponding with their authors. Others, however, become permanently unavailable as they languish as unpublished conference presentations or are relegated to filing cabinets and waste baskets. Hubbard and Vetter (1996) provide evidence of bias against replication studies in business journals. They report that fewer than 6% of 4,270 studies in the top business (including accounting) journals are replications. Armstrong and Pagell (2003, p. 8) note that ‘[i]t is very hard to obtain funds for replication alone, and journal editors prefer to publish new discoveries and to reject manuscripts that merely replicate previous findings’. Naturally researchers recognizing that their literature suffers from this survivorship bias would eschew MA for fear that MA would instill undue confidence in the generality of results. Although similar issues afflict other scientific disciplines, prior research suggests that publication bias is particularly acute in accounting. Surveying the empirical budgeting and control papers in Accounting, Organizations and Society, The Accounting Review and The Journal of Accounting Research from 1970 to 1987, Lindsay (1994) reports that 84.2% of the papers reject a null hypothesis and virtually none are replication studies. To the extent that those findings are representative of the accounting literature generally, MA studies focusing on only published research are likely to overstate the strength of treatment effects in the population and/or overstate the statistical significance of even weak results. Based on a citation analysis of seminal accounting research studies, Brown (1996) concludes that the accounting literature is severely under-cited, relative to the literatures of other disciplines, with even the most prolific authors attracting only about four annual citations on average. These results suggest that the accounting literature has very few progressive, developed research streams that would be amenable to MA. Publication bias is consistent with Bamber et al.’s thesis (2000) that the accounting research environment is restrictive, conferring hegemony on those who publish the first studies on given topics in the literature and erecting publication barriers to later studies reporting conflicting or non-significant results. We find, however, that even when we include all known studies of AC effectiveness to date, both published and unpublished, we cannot reliably aggregate across all of them because of significant inconsistencies in defining the construct ‘financial reporting quality’ and a relative absence of studies designed to replicate, and thereby enhance confidence in, the results of previous studies. Thus, we
320
B. Pomeroy & D. B. Thornton
turn to perverse researcher incentives as a second factor impeding MA in the accounting/auditing literature. 4.2. Researcher Incentives Whitley (2000) argues that researchers strive to produce innovative research to enhance their socially derived reputations. Innovation occasionally spawns seminal contributions but also constrains future researchers and interest groups to rely on the results of only a few, if any, comparable studies. In a similar vein, Burgstahler (1987) contends that successful researchers need to differentiate their studies from run-of-the-mill experiments by employing novel measures to proxy for broad concepts like FRQ, whether they undertake new studies or replication studies. Modeling accounting researchers’ expected utility, Burgstahler sees publications as major drivers of accounting researchers’ prestige and monetary rewards. He argues that the peer-review process generally filters out studies failing to reject null hypotheses and thus discourages even their submission to top-tier journals, since ‘. . . utility . . . depends primarily on the product of the reward to publication and the probability of publication’ (Burgstahler, 1987, p. 211). Thus, in the tradeoff between mitigating type I errors (i.e. rejecting a null hypothesis when the null is true) and type II errors (i.e. accepting a null hypothesis when the alternative is true), accounting researchers are motivated to focus unduly on the latter, so that published studies frequently exhibit high statistical power but low effective levels. MA, as it is customarily construed (i.e. statistically aggregating findings across an entire literature), is thus likely to yield unreliable results in literatures that exhibit substantial diversity in concept measurement; even comparing results informally across studies is a moot exercise. Consistent with this hypothesis, innovation in measuring FRQ in the AC literature severely inhibits our ability to apply MA to induce reliable implications of this research stream. Bamber et al. (2000, p. 123) allude to three aspects of the extant accounting research culture that could account for these findings, concluding that ‘. . . several facets of the practice of science in the academic accounting community combined to foster an environment where “the placement of the first research bricks affected the whole wall”’. First, individual researchers fail to stress how the limitations inherent in research-design choices affect inferences. Second, because the accounting research environment discourages replication studies, consumers of the research and authors of subsequent studies tend to make generalizations on the basis of only a few published studies. Third, in collectively failing to apprehend the impact of prior research-design choices, subsequent researchers tend to base their additional work on prior studies’ weakly supported inferences. Publication bias and perverse researcher incentives pose significant barriers to conducting MA studies in any literature. Our finding that the accounting literature contains a much lower proportion of published MA studies, even when compared
Meta-analysis and the Accounting Literature
321
with related business literatures, suggests that the biases Bamber et al. allude to are especially acute in the accounting literature.
4.3. Evidence Consistent with Publication Bias and Researcher Incentives Results in the AC independence literature are consistent with publication bias, with 13 of 20 published studies finding significant results. Moreover, the seven published studies finding non-significant results for AC independence deemphasize those results, focusing mainly on other relationships: Farber (2005) emphasizes findings suggesting that the discovery of fraud yields improvements in corporate governance; Ebrahim (2007) stresses associations between board characteristics other than AC independence and abnormal accruals; Lin et al. (2006) highlight a significant association between the size of ACs and earnings restatements; Rahman and Ali (2006) report that earnings management is positively related to board size; Yang and Krishnan (2005) discuss significant findings relating to AC governance expertise and AC tenure; Xie et al. (2003) focus on AC financial expertise and activity; and Agrawal and Chadha (2005) accentuate the importance of having an independent AC member with financial expertise. These observations are consistent with the accounting research environment fostering null-rejection bias that is likely to preclude MA reliability. Seven of the 27 papers we considered for inclusion in our MA remain unpublished, residing in electronic form on the SSRN. Although other factors, such as low research quality, could account for this observation, it is consistent with a first-mover effect in publication. Bamber et al. (2000) argue that the first paper on a given topic published in a major journal often lays ‘research bricks’. Authors of contemporary working papers on similar topics are then unlikely to invest the resources required to shepherd their manuscripts through the review process at top-tier journals because the likelihood of rejection is high once those bricks are in place. Our review of the AC independence literature suggests that only studies employing novel dependent variable measures are likely to penetrate this publication barrier. Despite the barriers to conducting MA identified above, at least two successful MA studies have appeared in the accounting and auditing literatures (Kinney and Martin, 1994; Hay et al., 2006). We believe that the success of these MA studies derives chiefly from the use of conceptually straightforward and consistent dependent variables (audit-related accounting adjustments and audit fees, respectively). With less precisely defined conceptual variables, such as FRQ, the potential for measurement diversity is high. This reduces the reliability of studies attempting to summarize or aggregate results in the literature. Even when conceptual variable measurement exhibits inconsistency across studies, however, our study demonstrates that MA can isolate and explore the effects of this diversity and examine how different groups of proxies for conceptual dependent variables are differentially related to the independent variable(s).
322
B. Pomeroy & D. B. Thornton
Thus, despite publication biases and perverse researcher incentives, quantitative review techniques can yield informative inferences. 5.
Conclusions and Implications
We conclude that it is not generally feasible for accounting and auditing researchers to employ meta-analyses to observe overall mean effect sizes except in a handful of cases where the variables examined exhibit consistency across the literature. However, we find that applying the standard quantitative review techniques of meta-analysis generates insights into the reasons for conflicting results in the literature concerning the association between audit committee independence and financial reporting quality and yields at least three preliminary conclusions and issues for future research to explore. First, the use of different financial reporting quality measures in the audit committee independence literature is responsible for more than half of the variation across studies in reported associations between audit committee independence and financial reporting quality proxies. This finding strongly suggests that extant proxies for financial reporting quality used in the audit committee independence literature are tapping into significantly different concepts. In particular, the impact of audit committee independence on high quality proxies, such as cumulative abnormal returns and the cost of debt financing, is likely to differ from its impact on proxies for egregiously low financial reporting quality, such as fraudrelated earnings restatements and SEC enforcement activities. Second, independent audit committees are more effective at enhancing audit quality (e.g. through averting auditor resignations) than they are at fostering financial statement quality (e.g. by making high quality accruals and avoiding restatements). For example, audit committee independence would actually reduce apparent financial statement quality if it identified the need for restatements and remedial, abnormal accruals. Third, financial statement quality and audit quality contribute to financial reporting quality in different, complementary ways. Additional theory is needed to understand financial reporting quality as a conceptual construct. Our summary of prior meta-analysis studies in accounting and our own experience in conducting a meta-analysis suggest that researchers’ perceived inability to perform reliable meta-analyses accounts for the dearth of such studies in the accounting literature. Although we grant that other areas of the literature could be more amenable to the application of traditional meta-analysis, and we encourage similar attempts by others interested in generalizing the results in those areas, examining the potential causes of this variability is likely to spawn more interesting conclusions than simply calculating overall mean effect sizes. A precondition for meta-analysis becoming more prominent in the accounting and auditing literature is awareness among researchers and journal editors of the barriers to performing meta-analysis and potential opportunities associated with it. We think that the present study has the potential to kindle this awareness. Awareness, however, is not a sufficient condition for alleviating the concerns we
Meta-analysis and the Accounting Literature
323
document. Empirical accounting and auditing research needs to focus, more than it has focused to date, on generating reproducible results that future researchers and today’s policy makers can rely on. Thus, referees and editors need to insist on researchers disclosing not just statistics but also effect sizes and explicit information relating to power (including the calculations that go into calculating power) so that others can readily construct sample sizes for replication or related research. Researchers also need to strive for consensus as to the conceptual variables used as proxies for research constructs like financial reporting quality. Acknowledgements For helpful comments on previous drafts of the paper, the authors are grateful to Tandy Chalmers, David Cooper, Yves Gendron, Michael Gibbins, Karim Jamal, Teri Shearer, Satyendra Singh and Jane Webster; to Marc Jegers and other participants at the 2006 European Accounting Association Annual Congress; and to the Editor and two anonymous referees. Appendix: Detailed Procedures and Results of Meta-analysis We are interested in the overall association between AC independence and FRQ, drawing on the entire extant literature. Therefore, we first summarize the reported results for each paper and calculate the mean correlation coefficient (r) effect size. After assessing each paper’s main results, we record the reported effect size or statistic. Since all of the papers report a statistic rather than directly reporting an effect size for the association, we need to convert the various statistical measures into correlation coefficient r effect sizes using the Lipsey and Wilson (2001) methodology. We then transform these r effect sizes into standardized (Zr) values and multiply the Zr values by their corresponding variance weight (n 2 3) to arrive at the mean r effect size. All calculations appear in Table A1. Converting Statistics to Effect Sizes The studies included within the MA employ a variety of test statistics to measure the association between AC independence and FRQ. These test statistics include t-statistics, chi-square, F-test scores, p-values and Z-statistics. To convert these statistics into r correlations, we apply the effect size formulas from Lipsey and Wilson (2001), which are as follows: t-statistic: chi-square statistic: F-statistic:
t ffi ESr ¼ pffiffiffiffiffiffiffiffi t2 þdf qffiffiffiffi 2 ESr ¼ xN
jESr j ¼
pffiffiffi F pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Fþn1 þn2 2
(A1) (A2) (A3)
324
Number 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Author Klein Xie et al. Be´dard et al. Williams Agrawal and Chadha Abbott et al. Yang and Krishnan Abbott et al. Lee et al. Farber Anderson et al. Bryan et al. Lennox Anderson et al. Felo et al. Wright Bradbury et al. Carcello and Neal Raghunandan and Rama
Reported stat. value
n
Effect size (ES )
ESZr
w
w ESZr (AC)
Standard error (SE)
w ES
w ES 2
6.11 20.41 0.02 23.02 0.82
692 282 300 465 318
20.09 20.02 20.13 20.14 20.01
20.09 20.02 20.14 20.14 20.01
689 279 297 462 315
264.96 26.81 240.13 264.64 24.03
0.04 0.06 0.06 0.05 0.06
264.77 26.81 239.89 264.22 24.03
6.09 0.17 5.36 8.93 0.05
6.01 0.93 11.60 23.67 0.67 22.38 2.22 24.61 0.02 21.29 2.83 0.01 11.62 0.05
176 896 156 380 78 1,052 1,291 457 1,241 119 64 252 374 124
20.18 0.03 20.27 20.19 0.08 20.07 20.06 20.21 20.07 20.12 20.32 20.16 20.18 20.18
20.19 0.03 20.28 20.19 0.08 20.07 20.06 20.21 20.07 20.12 20.34 20.16 20.18 20.18
173 893 153 377 75 1,049 1,288 454 1,238 116 61 249 371 121
232.34 27.78 242.81 270.56 5.68 276.92 279.44 297.25 284.31 213.72 220.53 239.29 266.10 221.40
0.08 0.03 0.08 0.05 0.12 0.03 0.03 0.05 0.03 0.09 0.13 0.06 0.05 0.09
231.97 27.77 241.72 269.75 5.67 276.79 279.34 295.79 284.18 213.65 219.79 238.97 265.41 221.18
5.91 0.86 11.38 12.90 0.43 5.62 4.89 20.21 5.72 1.61 6.42 6.10 11.53 3.71
B. Pomeroy & D. B. Thornton
Table A1. Calculations
20 21 22 23 24 25 26 27
Abbott et al. Carcello et al. Carcello and Neal Persons Parker et al. Ebrahim Lin et al. Rahman and Ali Totals 20.08 0.01 29.38
Confidence limits Negative interval Positive interval
20.10 20.07
492 254 223 222 161 2,360 212 194
Homogeneity test Q-Statistic Critical value Significance level
20.12 20.12 20.26 20.15 20.15 20.03 0.01 20.05
20.12 20.12 20.26 20.154 20.15 20.026 0.012 20.049
489 251 220 219 158 2,357 209 191 12,754
261.00 230.83 258.03 233.68 223.77 262.33 2.49 29.31
0.05 0.06 0.07 0.07 0.08 0.02 0.07 0.07 0.009
260.68 230.67 256.72 233.42 223.59 262.32 2.49 29.30 21,059.02
7.53 3.75 14.62 5.10 3.52 1.65 0.03 0.45 154.53
66.59 38.89 ,0.001 66.59
Note: The direction of the association (correlation) is coded as negative if the association is in the predicted direction and positive otherwise. We primarily consulted Lipsey and Wilson (2001) to calculate effect sizes and we used judgment, as recommended by Greenberg (1992), to code studies where the statistic used to test significance was not clear and for studies that employed firm-years rather than matched pairs; n ¼ sample size; w ¼ variance weight; Zr ¼ standardized correlation.
Meta-analysis and the Accounting Literature
Mean ES SE of mean ES Standardized (Z) ES
2.39 0.05 14.83 22.30 21.91 21.28 0.03 0.50
325
326
B. Pomeroy & D. B. Thornton
p-value:
first converted to t-statistic, then converted to r using equation (A1).
Computing the Mean Effect Size We measure the mean effect size r using the inverse weight variance method outlined by Lipsey and Wilson (2001). This technique standardizes the effect size r of each study as Z-values, as follows: 1þr ESZr ¼ 0:5 loge : 1r
(A4)
Standardizing the effect sizes as Z-values eliminates problems of the standard error formulation inherent in product – moment correlation. To incorporate the impact of sample size in the analysis, we compute the variance weight, wZr, of each study by subtraction (n 2 3), and multiply the weight by the standardized Z-value correlations. We then determine the mean Zr effect size using the following equation: P ESZr ¼
(wZr ESZr ) P : w Zr
(A5)
Finally, to interpret the results of this meta-analysis we convert the mean Zr effect size back into standard correlation form as follows:
ESr ¼
e2ESZr 1 e2ESZr þ 1
:
(A6)
To test for significance, we use the following formula from Lipsey and Wilson (2001), which assesses significance based upon the Z-value tested at the 0.05 significance level: z¼
jESZr j : SEESZ
(A7)
r
Test for Homogeneity of Variances We conduct a homogeneity analysis to determine whether the variety of proxies for FRQ actually captures the same construct (i.e. whether they have equal variances). The homogeneity test (Q) provided by Lipsey and Wilson (2001) is
Meta-analysis and the Accounting Literature
327
computed as follows:
Q¼
X
wZr ES2Zr
P 2 wZr ESZr P : w Zr
(A8)
The result of the Q-test is compared to a critical chi-square value, where the degrees of freedom are defined as the number of studies less one, to determine whether to reject the null hypothesis of homogeneity. Conclusion Across the literature examining the association between AC independence and FRQ we find that the probability is 0.95 that the true association (correlation) is between 0.07 and 0.10. We also find significant heterogeneity of variances across the measurements employed to proxy for FRQ. Notes 1
We defined an accounting journal as any publication with the word ‘accounting’ or ‘auditing’ in the journal title. 2 Brierley (1999), and Ahmed and Courtis (1999) appeared in the same issue of The British Accounting Review. 3 We do not attempt to assess whether papers are unpublished because of publication bias or inferior quality stemming from lack of power in the research design or inferior exposition. Superior quality of published studies is likely to be associated with authors presenting them at prestigious workshops (Brown, 2005). Publication bias is likely to be consistent simply with published papers being submitted to journals prior to similar, unpublished papers; however, we cannot locate reliable data on submission dates for unpublished studies. 4 Collins and Hribar (2002) outline significant pitfalls involved in estimating accruals from published financial data. 5 Detailed instructions for conducting MA beyond the purposes of this paper appear on the website http://mason.gmu.edu/dwilsonb/ma.html.
References
Denotes studies that were canvassed for meta-analysis. Abbott, L., Park, Y. and Parker, S. (2000) The effects of audit committee activity and independence on corporate fraud, Managerial Finance, 26(11), pp. 55–67. Abbott, L., Parker, S., Peters, G. and Raghunandan, K. (2003) The association between audit committee characteristics and audit fees, Auditing: A Journal of Practice and Theory, 22(2), pp. 17–32. Abbott, L., Parker, S. and Peters, G. (2004) Audit committee characteristics and restatements, Auditing: A Journal of Practice and Theory, 23(1), pp. 69– 87. Agrawal, A. and Chadha, S. (2005) Corporate governance and accounting scandals, Journal of Law and Economics, 48(2), pp. 371 –406. Ahmed, K. and Courtis, J. (1999) Associations between corporate characteristics and disclosure levels in annual reports: a meta-analysis, The British Accounting Review, 31(1), pp. 35– 61. Anderson, K., Deli, D. and Gillan, S. L. (2003) Boards of directors, audit committees, and the information content of earnings, Working Paper, University of Delaware.
328
B. Pomeroy & D. B. Thornton
Anderson, R., Mansi, S. and Reeb, D. (2004) Board characteristics, accounting report integrity, and the cost of debt, Journal of Accounting and Economics, 37(3), pp. 315 –342. Armstrong, J. and Pagell, R. (2003) The ombudsman: reaping benefits from management research: lessons from the forecasting principles project, Interfaces, 33(6), pp. 91–111. Bamber, L., Christensen, T. and Gaver, K. (2000) Do we really ‘know’ what we think we know? A case study of seminal research and its subsequent overgeneralization, Accounting, Organizations and Society, 25(2), pp. 103 –129. ´ Bedard, J., Chtourou, S. and Courteau, L. (2004) The effect of audit committee expertise, independence, and activity on aggressive earnings management, Auditing: A Journal of Practice and Theory, 23(2), pp. 13–35. Borkowski, S. (1996) An analysis (meta- and otherwise) of multinational transfer pricing research, The International Journal of Accounting, 31(1), pp. 39 –53. Bradbury, M., Mak, Y. and Tan, S. (2004) Board characteristics, audit committee characteristics and abnormal accruals, Working Paper, Massey University and National University of Singapore. Brierley, J. (1999) Accountant’s job satisfaction: a meta-analysis, The British Accounting Review, 31(1), pp. 63 –84. Brown, L. (1996) Influential accounting articles, individuals, Ph.D. granting institutions and faculties: a citational analysis, Accounting, Organizations and Society, 21(7), pp. 723 –754. Brown, L. (2005) The importance of circulating and presenting manuscripts: evidence from the accounting literature, The Accounting Review, 80(1), pp. 55 –83. Bryan, D., Liu, M. and Tiras, S. (2004) The influence of independent and effective audit committees on earnings quality, Working Paper, State University of New York and Louisiana State University. Burgstahler, D. (1987) Inference from empirical research, The Accounting Review, 62(1), pp. 203 –214. Carcello, J. and Neal, T. (2000) Audit committee composition and auditor reporting, The Accounting Review, 75(4), pp. 453–467. Carcello, J. and Neal, T. (2003) Audit committee characteristics and auditor dismissals following ‘new’ going-concern reports, The Accounting Review, 78(1), pp. 95–117. Carcello, J., Hermanson, D., Neal, T. and Riley, R., Jr. (2002) Board characteristics and audit fees, Contemporary Accounting Research, 19(3), pp. 365 –384. Collins, D. and Hribar, P. (2002) Errors in estimating accruals: implications for empirical research, Journal of Accounting Research, 40(1), pp. 105–134. Cooper, W., Ho, J., Hunter, J. and Rodgers, R. (1985) The impact of the foreign corrupt practices act on internal control practices, Journal of Accounting, Auditing and Finance, 9(1), pp. 22– 40. Dechow, P. and Dichev, I. (2002) The quality of accruals and earnings: the role of accrual estimation errors, The Accounting Review, 77(4), pp. 35–59. DeZoort, T., Hermanson, D., Archambeault, D. and Reed, S. (2002) Audit committee effectiveness: a synthesis of the empirical audit committee literature, Journal of Accounting Literature, 21, pp. 38–75. Dole, D. and Schroeder, R. (2001) The impact of various factors on the personality, job satisfaction and turnover intentions of professional accountants, Managerial Auditing Journal, 16(4), pp. 234 –245. Ebrahim, A. (2007) Earnings management and board activity: additional evidence, Review of Accounting and Finance, 6(1), pp. 42–58. Farber, D. (2005) Restoring trust after fraud: does corporate governance matter?, The Accounting Review, 80(2), pp. 539–561. Felo, A., Krishnamurthy, S. and Solieri, S. (2003) Audit committee characteristics and the perceived quality of financial reporting: an empirical analysis, Working Paper, Pennsylvania State University, SUNY at Binghamton and University of Scranton. Fletcher, J. (1995) An examination of the selectivity and market timing performance of UK unit trusts, Journal of Business Finance and Accounting, 22(1), pp. 143–156.
Meta-analysis and the Accounting Literature
329
Gibbins, M., McCracken, S. and Salterio, S. (2007) The chief financial officer’s perspective on auditor–client negotiations, Contemporary Accounting Research, 24(2), pp. 387 –422. Greenberg, P. (1992) The use of meta-analysis techniques in developing dissertation proposals: guidelines to aid supervising faculty, Issues in Accounting Education, 7(2), pp. 145 –163. Greenberg, P., Greenberg, R. and Nouri, H. (1994) Participative budgeting: a meta-analytic examination of methodological moderators, Journal of Accounting Literature, 13, pp. 117– 141. Hay, D., Knetchel, W. and Wong, N. (2006) Audit fees: a meta-analysis of the effect of supply and demand attributes, Contemporary Accounting Research, 23(1), pp. 141–191. Hubbard, R. and Vetter, D. (1997) An empirical comparison of published replication research in accounting, economics, finance, management and marketing, Journal of Business Research, 35(2), pp. 153 –164. Kinney, W., Jr. and Martin, R. (1994) Does auditing reduce bias in financial reporting? A review of audit-related adjustment studies, Auditing: A Journal of Practice and Theory, 13(1), pp. 149–155. Klein, A. (2002) Audit committee, board of director characteristics, and earnings management, Journal of Accounting and Economics, 33(3), pp. 375–400. Lee, H., Mande, V. and Ortman, R. (2004) The effect of audit committee and board of director independence on auditor resignation, Auditing: A Journal of Practice and Theory, 23(2), pp. 131–146. LeLorier, J., Gregoire, G., Benhaddad, A., Lapierre, J. and Derderian, F. (1997) Discrepancies between meta-analyses and subsequent large randomized controlled trials, New England Journal of Medicine, 337(8), pp. 536 –542. Lennox, C. S. (2002) Opinion shopping, audit firm dismissals, and audit committees, Working Paper, Hong Kong University of Science and Technology. Levitt, A. (1998) ‘The numbers game’, remarks delivered at the NYU Center for Law and Business, New York, 28 September. Levitt, A. (2000) Remarks delivered at the Economic Club of Washington, Washington, DC, 6 April. Lin, J., Li, J. and Yang, J. (2006) The effect of audit committee performance on earnings quality, Managerial Auditing Journal, 21(9), pp. 921 –933. Lindsay, R. (1994) Publication systems biases associated with the statistical testing paradigm, Contemporary Accounting Research, 11(1), pp. 33–57. Lipsey, W. and Wilson, D. (2001) Practical Meta Analysis (Thousand Oaks, CA: Sage). Parker, S., Peters, G. and Turetsky, H. (2005) Corporate governance factors and auditor going concern assessments, Review of Accounting and Finance, 4(3), pp. 5– 29. Persons, O. (2005) The relation between the new corporate governance rules and the likelihood of financial statement fraud, Review of Accounting and Finance, 4(2), pp. 125–148. Peterson, I. (1989) Publication bias: looking for missing data, Science News, 7 January. Raghunandan, K. and Rama, D. (2003) Audit committee composition and shareholder actions: evidence from voting on auditor ratification, Auditing: A Journal of Practice and Theory, 22(2), pp. 253 –263. Rahman, R. and Ali, F. (2006) Board, audit committee, culture and earnings management: Malaysian evidence, Managerial Auditing Journal, 21(7), pp. 783 –804. Smith, M. and Glass, G. (1977) Meta-analysis of psychotherapy outcome studies, American Psychologist, 32(9), pp. 752–760. Thornton, D. (1994) What does it mean? What next? Commentary on and meta-analysis of the CAAA Education Project, Contemporary Accounting Research, Special Education Research Issue, pp. 179–185. Trotman, K. and Wood, R. (1991) A meta-analysis of studies on internal control judgments, Journal of Accounting Research, 29(1), pp. 180–192. Whitley, R. (2000) The Intellectual and Social Organization of the Sciences, 2nd edn (Oxford: Oxford University Press). Williams, S. (2002) Association between earnings management and properties of independent audit committee members, Working Paper, Singapore Management University.
330
B. Pomeroy & D. B. Thornton
Wright, D. (1996) Evidence on the relation between corporate governance characteristics and the quality of financial reporting, Working Paper, University of Michigan. Xie, B., Davidson, W. and DaDalt, P. (2003) Earnings management and corporate governance: the role of the board and the audit committee, Journal of Corporate Finance, 9(3), pp. 295–316. Yang, J. and Krishnan, J. (2005) Audit committees and quarterly earnings management, International Journal of Auditing, 9, pp. 201– 219. Zager, M., McGaha, J. and Garcia, L. (2001) Prison privatization and public budgeting: a metaanalysis of the literature, Journal of Public Budgeting, Accounting and Financial Management, 13(2), pp. 222 –244.
