Corporations are increasingly relying on cross-cultural train- ing of expatriate managers to reduce personnel costs. Previous research has shown consistent ...
International Journal of Training and Development 5:2 ISSN 1360-3736
A meta-analysis of the effects of cross-cultural training on expatriate performance and adjustment Mark A Morris and Chet Robie Corporations are increasingly relying on cross-cultural training of expatriate managers to reduce personnel costs. Previous research has shown consistent validity for different training methods. This meta-analysis investigated the available empirical literature, identifying 16 studies for expatriate adjustment (total n = 2270) and 25 studies for expatriate performance (total n = 2490). Mean effect sizes were lower than in a previously published meta-analysis and were significantly different from zero for both performance (r = .26) and adjustment (r = .13). Until the effects of potential moderators are better understood, prescriptions for cross-cultural training design should be made cautiously.
Introduction Justifications for the use of cross-cultural training (CCT) are not difficult to find with more than half a trillion dollars in direct American investment in foreign markets (United States Department of Commerce, 1994). Some observers (Tung, 1982) have noted an increase in expatriate managers to oversee the multinational organisations’ burgeoning investment. Direct costs of sending an expatriate overseas can be as high as $220,000 (Birdseye and Hill, 1995), and may cost American business more than $2 billion annually. Turnover is also a serious and growing problem, with between 16 per cent and 40 per cent returning early (Black, 1988; Rahim, 1983) from their assignments (compare this to the 5 per cent domestic rate). Cost estimates vary, but all researchers agree that bringing in an expatriate manager is an expensive ❒ Mark A Morris, Department of Psychology, University of Houston, Houston, TX 77204–5341, USA. Chet Robie is at the College of Business Administration, Niagara University, Lewiston, NY 14109, USA. Blackwell Publishers Ltd. 2001, 108 Cowley Road, Oxford OX4 1JF, UK and 350 Main St., Malden, MA 02148, USA.
112 International Journal of Training and Development
proposition. Estimates of failure rates vary from 20 per cent (Black and Mendenhall, 1990) to as much as 85 per cent (Walton, 1990). If expatriates are expensive and risky, but the organisation is committed to having American managers in place at an overseas location, what can be done to minimise the costs and risks associated with such a placement? Many researchers have suggested that training is the answer (Black and Mendenhall, 1990; Landis and Brislin, 1983; Kealey and Protheroe, 1996). However, a frequently cited survey (Tung, 1981) reported that only 32 per cent of American multinational corporations had formal training programs in place for expatriates. An Andersen Consulting survey (Cuthill, 1997) of 32 Fortune 500 companies identified as Best of Class revealed that 94 per cent of these multinational firms offered at least a language training program for international assignees and 69 per cent offered some additional form of cross-cultural training. This suggests that CCT may be gaining more credence from top executives as an effective developmental tool for soon-to-be expatriates. More and more training is moving to a web-based medium, as training organisations take advantage of the anytime/anywhere accessibility of the Internet. Andersen Consulting has partnered with TerraCognita.com to create CultureSavvy.com, a site integrating video and text-based training in structured modules designed to build cross-cultural skills in expatriates. These modules focus on communication skills, adjustment (for the entire expatriate family), and re-entry issues. Their flat fee pricing model is likely to be more appealing to large organisations. See Appendix B for an annotated list of web sites of other major players in online CCT. Most programs typically last 1–5 days, are region-specific, involve the entire expatriate family, and are delivered sometime between 3 months and 3 weeks prior to departure. No data on the effectiveness of these training programs was reported by the organisations listed in Appendix B—they are listed solely for the purpose of providing background on the current state of the market for expatriate training. Responses from senior human resource and international relocation professionals to a recent survey (of over 250 organisations from all industries, including high technology firms) indicate that expatriation rates continue to climb overall, although some areas are seeing less international assignees (Windham International, 1999). According to the survey, 41 per cent of corporate revenues are generated overseas. Additional findings were that 63 per cent of companies provide cross-cultural training; 70 per cent of their expatriates participate when it is available, and 81 per cent of respondents rate this preparation as having great or high value. In the influential Kirkpatrick model of training evaluation (Kirkpatrick, 1998), this type of trainee attitude survey is considered merely a reaction measure. While it assessed the face validity of the training, it did not directly assess the amount of information learned or determine the extent to which newly acquired behaviors would generalise to the actual work setting. Most of all, surveys of participant reactions do not indicate the bottom-line results in improved performance, productivity and profits that can be expected from the training. In short, this rosy view of intercultural training may at best be premature or at worst, unwarranted by the available data. Several field studies were published in the early 1990s reporting negative or trivial validity for CCT in predicting job performance (Black and Gregersen, 1991b; Hammer and Martin, 1992; Cui and Awa, 1992). This seeming contradiction between common sense (and common practice) and research findings only becomes more evident when the full range of results is examined. Different studies have obtained correlations between CCT and expatriate adjustment ranging from −.42 (Black, 1988) to .57 (Earley, 1987), while correlations between CCT and job performance range from −.08 (Black and Gregersen, 1991b) to .79 (Earley, 1994). When a wide range of effect sizes abounds in the literature, metaanalytic techniques may be applied to clarify the matter by correcting for the effects of artifacts such as sampling error, restriction of range, and criterion unreliability (Hunter and Schmidt, 1990). Meta-analysis is a method developed in the late 1970s to summarise and integrate research findings from multiple articles (Glass, 1976). By the 1980s, the initial work Blackwell Publishers Ltd. 2001.
Meta-analysis of cross-cultural training 113
had been refined and extended, and different factions began to emerge, as a number of seminal books were published (Hedges and Olkin, 1985; Hunter, Schmidt and Jackson, 1982; Rosenthal, 1980). The goal of meta-analysis is to resolve conflicting findings of multiple studies on the same topic by combining their results in a systematic fashion. After combination, the resulting effect size is based on a much larger sample size because it contains cases from multiple studies, thereby increasing the statistical power and generalisability to multiple contexts. Furthermore, using metaanalytic techniques, the researcher can control for the effects of sampling error, unreliability, and restriction of range. After removing the effects of these artifacts, the researcher can compare the observed variability among the effect sizes in a given data set to the amount of variability expected due to sampling error and other artifacts. If most of the observed variability in effect sizes (at least 75 per cent) can be attributed to a combination of sampling error, restriction of range, and unreliability, the researcher can use the mean corrected correlation or effect size derived from the meta-analysis, and validity is said to generalise (Hunter and Schmidt, 1990). Validity generalisation is the goal of any researcher who wishes to know with some degree of confidence if a particular relationship or effect is stable and reliable across different types of samples, occupational groups and environments.
Background to cross-cultural training evaluations In the most often-cited narrative review of the effectiveness of CCT (Black and Mendenhall, 1990), the authors proposed a theoretical framework in the field of crosscultural training by linking principles of social learning theory to training effectiveness in an international environment. Other researchers (Parker and McEvoy, 1993) have suggested a mediating role for adjustment between training and performance effectiveness. A previous meta-analysis (Deshpande and Viswesvaran, 1992) examined the effect of cross-cultural training on the criteria of job performance and adjustment. This study found moderately strong correlations (corrected for sampling error and measurement error) of .39 and .43 for the effects of CCT on performance and adjustability respectively (Deshpande and Viswesvaran, 1992). The present study represents an extension of previous work in several important ways. First, a more comprehensive literature review was done, resulting in 78 empirical studies, 19 of which were published after the earlier meta-analysis. Thirty-one effect sizes drawn from these studies (9 of which were published after the earlier meta-analytic review) were able to be coded for either the performance or adjustment construct. The median year published of the studies in our database is 1986, as compared to 1982 for the studies in the previous meta-analysis. On the criterion side, more specific levels of adjustment were examined, including measures of stress (Befus, 1986) and work adjustment rather than general adjustment (Black, 1988; Black and Gregersen, 1991b). The earlier authors also used a variety of performance criteria, including early return from assignment, and ratings of intercultural communication. Based on a well-accepted definition of competent intercultural behavior (Landis and Brislin, 1983), new criteria for successful performance were added, including perceptions of cultural competence, awareness of cultural differences (Gannon and Poon, 1997) and technical knowledge about another culture (Hammer and Martin, 1992). The meta-analysis conducted in the present study also differed along some important judgmental dimensions. Subjectivity can contaminate the meta-analytic process, particularly regarding the difficulty in making judgement calls regarding the construct measured in a particular study (Wanous, Sullivan and Malinak, 1989). These judgement calls seem particularly relevant when considering the differences between the selection criteria in the current study and in the prior study (Deshpande and Viswesvaran, 1992). Some articles in the 1992 meta-analysis empirically examined the effectiveness of intercultural training on different ethnic groups within America’s borders—an important topic to be sure—but perhaps not sufficiently similar to pre-departure training for expatriates to justify aggregation. In addition to more recent studies, some of the studies included in earlier meta-analysis were not selected 114 International Journal of Training and Development
Blackwell Publishers Ltd. 2001.
for inclusion in the present study. Possibly to bolster a relatively small sample size, the authors included several studies that involved intercultural training programs within the United States or involved racial sensitivity training (Deshpande and Viswesvaran, 1992). As these educational programs were not composed of expatriates or include a non-American culture in which to collect criterion information, they did not meet the criteria for inclusion in the current study. These studies may be useful in determining the effectiveness of cross-cultural training within America’s borders or between American ethnic groups, but our judgement was that the fidelity of the study would suffer if they were to be aggregated with studies of expatriate managers working abroad. The purpose, sample, and criterion variables are all quite different between the two sets of studies. The purpose of the current study is to explore the relationship between CCT and the performance and adjustment constructs and to reveal the underlying state of affairs, after controlling for artifacts and examining moderators. It is expected that CCT will continue to be related to adjustment and performance, and therefore have implications for the costs and benefits associated with expatriate assignment. Hypothesis 1: CCT will be significantly and positively related to both performance and adjustment of expatriates. Training issues Citing the high costs of failure for American executives assigned abroad, some authors (Black and Mendenhall, 1990) have embraced cross-cultural training as the answer for international organisations. Cross-cultural training was defined as ‘those educative processes that are designed to promote intercultural learning, by which we mean the acquisition of behavioral, cognitive and affective competencies associated with effective interaction across cultures ‘ (Landis and Brislin, 1983). As the global village becomes a marketplace, the call for cross-cultural training continues to resound in the business and academic world (Thomas, 1996), but despite advances in training design in other areas, specific strategies for improving the performance of expatriate managers remain elusive. Despite their competence at home, expatriates may be unable to transfer their successful managerial practices abroad (Tung, 1987). Transfer of training is particularly meaningful for the organisations that invest heavily in an expatriate. Different organisations may have different tolerances for deviations from organisational norms. Effective training programs allow trainees to recognise and apply rules for what is not within the normal behavioral range at a particular organisation (Goldstein, 1993). Cross-cultural issues A frequent choice of theoretical framework for cross-cultural researchers is Hofstede’s model (Hofstede, 1980). This framework is built on the belief that all cultures could be classified as a function of four dimensions: power distance, individualism– collectivism, masculinity–femininity, and uncertainty avoidance. Many cross-cultural study results are interpreted within this view, as it has implications for the development of training interventions and intercultural adjustment. Methods of cross-cultural training are replete with Hofstede’s influence and elements of cognitive psychology. They can generally be placed in one of three categories: cultural awareness, attribution training, and the scenario-based cultural assimilator (Fiedler, Mitchell and Triandis, 1971). Cultural awareness training involved teaching the trainee to identify components unique to their own culture, thereby heightening awareness of analogous components in other cultures. Once trainees begin noticing culturally specific components, they can attend and respond to cultural stimuli appropriately. If they remain oblivious to the culture-specific aspects of their Blackwell Publishers Ltd. 2001.
Meta-analysis of cross-cultural training 115
environment, their chances of successful adaptation will decrease significantly. Language training would be considered part of this category as trainees learn grammar and vocabulary that are analogous to those of their native language. Attribution training is based on a cognitive model of behavior. The objective is for trainees to develop internal cognitive structures that allow them to interpret situations and make attributions as though the trainees were members of the culture in question. This ability to analyse social situations from the perspective of a native is thought to be invaluable in successful adjustment and performance. Behavioral modeling may fall into this category if the trainee is expected to infer or induce underlying interpretive mechanisms from observation of a model or example of behavior. The cultural assimilator technique has been the focus of considerable research (Becker, 1982; Bhawuk, 1998; Brislin, 1986; Cushner, 1989; Harrison, 1992; Landis, Day, McGrew, Thomas and Miller, 1976; Symonds, O’Brien, Vidmar and Hornik, 1967; Tolbert and McLean, 1995; Weldon, Carlston, Rissman, Slobodin and Triandis, 1975; Worchel and Mitchell, 1972; Yarbro and Lynette, 1988). The approach of the culture assimilator is to provide scenarios of cross-cultural situations requiring a decision by the expatriate, then offer four decision alternatives. The trainee selects an option and defends his/her choice. If the choice is incorrect, the trainee is told why that choice is an inappropriate response and asked to select again. The reasons behind the choice are more important than the option selected as they indicate the schemata used by the trainee to make inferences about culture-appropriate behavior. The goal of the cultural assimilator is to provide the trainee with experience making decisions in realistic social situations, much like a role-play, while imparting knowledge about the behavioral norms specific to that culture. As each training format differs in terms of theory and application, it is expected that the type of training will be related to effectiveness of the training program for the expatriates. In addition, field studies may be expected to obtain lower effect sizes than lab studies because of the strength of the manipulation and measurement difficulties. Another moderator that has been tested as a form of quality control for file drawer effects is the publication status of a study. Due to publication bias toward studies with significant results, unpublished studies usually have lower effect sizes. Hypothesis 2—Type of training will moderate the relationship between CCT and the dependent variables of adjustment and performance. Hypothesis 3—Study setting (lab vs. field) and publication status will moderate the relationship between CCT and the dependent variables of performance and adjustment.
Method Article search An attempt was made to acquire published and unpublished articles and studies relating to cross-cultural training effectiveness. Authors were contacted via telephone and email and multiple keyword searches using synonyms and wildcard operators were conducted on PsycLit, PsycInfo, ABI/Inform, Dissertation Abstracts and ERIC. Keywords included all possible derivatives and combinations of the following terms: cross-cultural, training, adjustment, performance, effectiveness, culture, expatriate, sojourner, job, work, abroad, intercultural, international, foreign, American. In addition, reference lists of all articles, including previous narrative and meta-analytic reviews were scoured and potential articles were reviewed for inclusion. Other tactics for obtaining articles included: posting messages in newsgroups and other online forums, and contacting private sector organisations for in-house research documents. The Social Science Citation Index was used to identify all authors citing eleven of the most prominent articles in the field. Some (Black and Mendenhall, 1990; Church, 1982) were selected because they were among the most frequently cited and 116 International Journal of Training and Development
Blackwell Publishers Ltd. 2001.
influential articles in the field of performance effectiveness and adjustment respectively, while another (Deshpande and Viswesvaran, 1992) was the most recently published meta-analysis on the topic. The remaining articles constituted a representative sample of the recent empirical literature on cross-cultural training effectiveness. Selection criteria Due to our focus on the relationship between training and performance and/or adjustment of American expatriates, we did not include studies focusing on adjustment of repatriates (returning expatriates), citizens of other countries, or expatriate adjustment/performance issues that were related to selection strategies rather than training methods. Also, the rich literature on cross-cultural training and adjustment/performance within American subcultures was not included. In order to be selected for inclusion, relevant studies also had to report unique quantitative data regarding the variables of interest. If multiple publications were based on the same data, then only one of those studies was included in the analysis. Dependent variables Consistent with earlier work (Deshpande and Viswesvaran, 1992), variables that were considered to be performance criteria for the purposes of this study included: ratings of cross-cultural skill development, rate of early return from overseas assignment, and supervisor ratings of job performance. Adjustment, originally defined as ‘the process of achieving harmony between the individual and the environment’ (Hannigan, 1990) included work adjustment (when available), or general adjustment. Studies were coded independently by four individuals trained in meta-analytic methods. Two were graduate students in industrial organisational psychology, one was a post-baccalaureate student taking a graduate-level course in meta-analysis and one was a volunteer coder who possessed relevant work experience. All coders became familiar with meta-analytic methods through reading important articles and practiced coding together before creating independent files. Agreement was reached in 95 per cent of all possible cases, and the remaining 5 per cent were resolved through mutual discussion.
Results In some cases, mean effect size was used when multiple dependent variables were presented as tapping the performance or adjustment construct domain. According to previous research (Rosenthal and Rubin, 1986), this results in a conservative estimate of effect size, so the mean validity estimate resulting from this study should be interpreted accordingly. Effect sizes were calculated for some studies based on the frequently cited formula for Cohen’s d (Hunter and Schmidt, 1990; Rosenthal and Rubin, 1986). When only means and standard deviations were available for a study, these were converted to d and finally to r-values through Kenny’s Meta software (Kenny, 1997). To control for unreliability in the criterion as well as sampling error, an artifact distribution was computed using available data. Reliability data were available for 6 out of 16 studies investigating adjustment and for 11 out of 25 studies investigating job performance. This was deemed sufficient to compute the artifact distribution (Hunter and Schmidt, 1990). Information on range restriction was not available. Data were analysed using Schwarzer’s software (Schwarzer, 1996). As is evident from Table 1 for the dependent variable of adjustment, the 90 per cent confidence interval does not include zero, and so the mean validity coefficient, expressed as an r value, (raw = .12, corrected =.13) is considered significantly different from zero. The lower bound of the 90 per cent credibility interval, which controls for artifactual Blackwell Publishers Ltd. 2001.
Meta-analysis of cross-cultural training 117
Table 1: Meta-analytic results for the effects of cross-cultural training on adjustment Correcting Correcting for for sampling sampling error error and unreliability Mean r Percentage of Total Var acct for by artifacts Mean standardized difference (g) Confidence Interval (95 per cent) Upper bound Lower bound Fail-safe n for critical r of .05
Outliers removed correcting for all artifacts
0.126 12.40
0.135 12.42
.19 111.84
.26
.27
.39
0.17 0.08 24.28
0.59 −.33 27.05
.19 .19 30.91
Note: k = 16 studies, total n = 2270. Outlier data were from k = 11 studies, total n = 1452. All data have been transformed using Fisher’s Z with Schwarzer’s meta-analytic software. Artifact corrections used the Schmidt and Hunter method.
variance (Whitener, 1990) does in fact include zero. The percentage of variance accounted for by artifacts is only 12.62 per cent (corrected), far below the standard 75 per cent threshold (Hunter and Schmidt, 1990). Also, the 2 value is significant at p ⬍ .001. Therefore, moderators may be implicated as the culprits explaining the remaining variability. In Table 2, the mean validity coefficient for job performance is larger in an absolute sense (.26 corrected, .23 uncorrected) than was the coefficient for adjustment. Once again, the 95 per cent confidence interval does not include zero in either condition, suggesting that the population parameter is nonzero. However, similar to the effect found for adjustment, the 90 per cent credibility interval does include zero. The percentage of variance accounted for by artifacts is only 15.75 per cent (corrected), well below the 75 per cent threshold (Hunter and Schmidt, 1990). Also, the 2 value is significant at p ⬍ .001. These results strongly argue for the further investigation of Table 2: Meta-analytic results for the effects of cross-cultural training on performance Correcting Correcting for for sampling sampling error error and unreliability Mean r Percentage of Total Var acct for by artifacts Mean standardized difference (g) Confidence Interval (95 per cent) Upper bound Lower bound Fail-safe n for critical r of .05
0.126 14.76
Outliers removed correcting for all artifacts
0.27 15.16
.13 38.21
.53
.58
.26
0.29 0.21 101
0.74 −.20 110.73
.40 −.15 27.93
Note: k = 25 studies, total n = 2490. Outlier data were from k = 18, total n = 1748. All data have been transformed using Fisher’s Z with Schwarzer’s meta-analytic software. Artifact corrections used the Schmidt and Hunter method.
118 International Journal of Training and Development
Blackwell Publishers Ltd. 2001.
moderators as they are likely to be soaking up some of the remaining observed variance. Taken together, these results support Hypothesis 1. The recommended Fisher’s Z transformation was used to normalise the r distribution (Hedges and Olkin, 1985). In all cases, Z values were significant at p ⬍ .01. The average effect size for each dependent variable should be interpreted in light of Cohen’s rule for effect sizes, which considers .2 to be a small effect (Kenny, 1987). In both cases, the average effect size is significantly different from zero. The fail-safe n represents the number of studies with zero effect size that there would have to be to make the result no longer statistically significant (p 05, two tailed). According to this measure (Orwin, 1983), it would require more than 100 studies with an effect size of zero to make the result for the job performance dependent variable non-significant but only 26 studies to classify the adjustment dependent variable as non-significant. The Chi-square tests of homogeneity indicate that the effect sizes in these studies differ significantly, based on factors other than sampling error, again indicating that a search for moderators is warranted. Outlier analysis A SAM-D outlier analysis was computed on the individual estimates of effect size and the resulting data file was also meta-analysed in the Schwarzer software (Schwarzer, 1996; Stajkovic, 1999). For this analysis, all r-values were converted to effect sizes using standard methods (Hedges and Olkin, 1985). Using the Qt homogeneity statistic (compared to a 2 critical value of 10.85 for alpha = .05 for k = 20), the following studies were iteratively removed from the analysis for the dependent variable of performance: (Earley, 1994) with a Qt of 52; (McDaniel et al., 1988) with a Qt of 26; (Chemers, 1969) with a Qt of 25; (Lefley, 1985) with a Qt of 17; (Earley, 1987) with a Qt of 16.7; (Earley, 1994) Study II with a Qt of 16.8; and (Lightner, 1993) with a Qt of 14. This resulted in a sample of 1,748 individuals from 18 studies. An analysis resulted in a corrected mean r = .13, with about 38 per cent of the variance accounted for by artifacts (see Table 2). In the analysis of the dependent variable of adjustment, five studies were iteratively removed based on their Qt values which exceeded the 2 critical value of 4.58 for k = 11. They included: (Black and Gregersen, 1991a) with a Qt of 44; (Earley, 1987) with a Qt of 29; (Lefley, 1996) with a Qt of 26; (Black, 1988) with a Qt of 22; and (Black and Gregersen, 1991b), with a Qt of 9. This resulted in a sample of 1452 individuals from 11 studies. Schwarzer’s software (Schwarzer, 1996) reported a mean r = .19 with about 11 per cent of the variance accounted for by artifacts (see Table 1). Moderator analysis Moderator analyses were performed using ANOVA on the weighted data (Kenny, 1987). Hypothesis 2 could not be tested because of the small sample of studies available. For the dependent variable of job performance, ten studies (total n = 641) out of the 32 used to investigate at least one of the dependent variables were coded as lab studies. No significant differences were found for the effect sizes of lab vs. field studies (p ⬎ .72), but significant differences were found for sample sizes (p ⬍ .01) with field studies averaging more than twice the sample size of lab studies. When the variable of publication status was examined as a moderator of effect size, only three studies were coded as unpublished (n = 283), and the resulting ANOVA was NS, p = .068, despite the presence of mean differences in the effect sizes (.31 vs. .21). Given the small number of unpublished studies it is reasonable to conclude that with more power, this effect might well have been significant. No statistically significant differences in sample sizes were identified between the published and unpublished groups. For the dependent variable of adjustment, only two studies (n = 195) were identified as unpublished. When weighted cases were examined, the effect sizes were not significantly different between groups (p ⬎ .23) despite mean differences that may Blackwell Publishers Ltd. 2001.
Meta-analysis of cross-cultural training 119
indicate low power. Sample sizes were significantly different (p ⬍ .02), with published studies having significantly larger sample sizes. Three studies (n = 318) were coded as lab studies examining the effect of CCT on adjustment. Effect sizes were significantly different for the two groups. Lab studies averaged a larger effect size (.38 vs. .11). Thus, partial support for Hypothesis 3 was found in this limited moderator analysis.
Discussion Although cross-cultural training is widely prescribed for expatriates, the results of this meta-analysis show the effectiveness of CCT somewhat weaker than expected and can vary widely. Unfortunately, few organisations systematically evaluate or validate the effectiveness of their training programs and fewer still are available to the public. The relatively small number of published studies often involve a mixture of different training methods. For example, a single class may use lecture format, videotapes and practice ratings with feedback at different times during the course. This makes it difficult to confidently estimate the effect of moderators such as type or method of training, items of singular interest to developers of cross-cultural training. The central findings of this study are the mean validity coefficients for performance (.26) and adjustment (.12) and the limits to their interpretation. Because of the small amount of variance accounted for by sampling error and criterion unreliability, the validity cannot be considered to generalize to a wide domain of jobs. This may be partly due to the immense diversity in cultures faced by expatriates and the interaction between individual differences of expatriates and the work environments to which they are assigned. The comfortably large fail-safe n of more than 100 allows the reader to feel reasonably secure in concluding that the relationship between crosscultural training and job performance is positive and non-zero. Also of interest were the results of the analysis of the datasets modified by removal of statistical outliers (Stajkovic, 1999). Although the sample size was reduced, the data for adjustment revealed a larger and more robust effect, with 100 per cent of the variance accounted for by artifacts, resulting in confidence intervals that were infinitely small. Conversely, the mean correlation for the dependent variable of performance declined significantly, from .26 to .13 when the outliers were removed from the data. Although the variance accounted for by artifacts more than doubled, to 38.21 per cent, the failsafe n was reduced due to the lower sample size. Complete results can be seen in Table 1 and Table 2 for performance and adjustment, respectively. There were several differences relative to the most recent meta-analysis, most likely due to the strategy used to select and define the dependent variables. Unlike an earlier meta-analysis (Deshpande and Viswesvaran, 1992), the 90 per cent credibility interval did include zero for both dependent variables. This was found despite a much larger sample size (at least 1500 more subjects for each dependent variable). Mean correlations were also much lower in this study than in the previous metaanalysis. Adjustment dropped from .43 to about .13. Performance fell from a robust .39 to around .26. The fail-safe n was larger in the current study, indicating more confidence in the results. For adjustment, the fail-safe n was 26 in the current study, up from 8 in prior research (Deshpande and Viswesvaran, 1992). The fail-safe n for performance went from 7 in 1992 to 110 in the current study, probably due to the larger sample size. As in the current study, earlier researchers were able to account for only a small amount of variance (21 per cent for adjustment, 34 per cent for performance), falling far short of the 75 per cent needed for validity to generalise, and once more underscoring the importance of a comprehensive moderator analysis. Two moderators were examined, publication status, and study setting. Field studies consistently had larger sample sizes than lab studies regardless of whether the dependent variable was adjustment or performance. Study setting significantly moderated the relationship between CCT and adjustment, with lab studies enjoying effect sizes that were almost three times as large as those reported in field studies. This finding was not replicated for job performance as p ⬎ .07 NS. Bootstrapping or some 120 International Journal of Training and Development
Blackwell Publishers Ltd. 2001.
other resampling method may be indicated here as the mean effect size differences suggest that low power may have affected this result. The current study supports the use of cross-cultural training for expatriate managers, with the caution that the effectiveness of these programs should be carefully evaluated. The high costs and increasing frequency of expatriate assignments have led to wide application of CCT, perhaps resulting in non-systematic training methods, compressed time frames, and uncertain benefits. All training programs should be systematically developed, based on a needs analysis, and rigorously evaluated in terms of both the affective responses of trainees, measures of learning and knowledge, and actual turnover rates and cultural competence evaluations as well as performance and adjustment of expatriates. Policy-makers should build in evaluation systems to their CCT programs to ensure that the programs attain results. As this research shows, effects of CCT can be as diverse as the countries to which expatriates are assigned. Limitations of the study include the inability to investigate some other potentially important moderators due to the difficulty of converting the data from some studies to effect sizes or because too few empirical studies reported information on them. These moderators should be examined in future analyses. Research support exists for examining the following moderators: training type/method, time in training, total number of trainees, training content and source of rating (Landis and Brislin, 1983; Kealey and Protheroe, 1996). A consistent limitation encountered in the course of this study was the dearth of empirical research. More than 220 published and unpublished studies were identified that addressed issues related to cross-cultural training’s impact on adjustment or performance. Of these, only 31 reported data that could be coded or transformed into an effect size (some studies reported data on both performance and adjustment and some articles contained more than one effect size relating to the same constructs). Most of the extant literature consists of the anecdotal experiences of former expatriate managers and tends to focus on rules of thumb or broad guidelines for behavior and training design without empirical support. A theoretical model of the relationships between CCT, adjustment and performance, intent to leave and turnover could be also developed. Some researchers (Birdseye and Hill, 1995) have already described some of the relationships between adjustment, intent to leave, and actual turnover. If researchers can begin to provide guidelines for practitioners on how to structure and design CCT for optimal efficiency, organisations may be able to derive more consistent benefits from such pre-departure training programs. Guidelines for training design have been developed for many different forms of training (Goldstein, 1993), and are usually based on a needs assessment and the resulting instructional objectives. The results of this study suggest that such traditional approaches may underestimate the complex interactional dynamic involved in global business patterns. Expatriate managers are asked to perform complex jobs, requiring many skills (Tung, 1981) and to work effectively across multiple cultural contexts, sometimes on the same day or even the same hour. The diverse context and unique interactional patterns of expatriate managers may require new predictor variables or different training methodologies to adequately prepare the employee. Existing training systems advertised on the Internet have yet to show empirical results to support their utility. Theoretical specifications could also benefit from a structural modeling perspective of the nature of the impact of training programs on performance and adjustment while accounting for the role of the spouse and different types of turnover. In addition, selection issues were not tested in this analysis. Few studies have examined the flow of decedents and antecedents in a cross-cultural training context. Logically, it would be expected that variables on which individuals differ would be related to success on international assignments. Therefore, a plausible model might include cognitive ability, personality factors (such as sensation seeking or tolerance for ambiguity), biodata, vocational interests, work experiences, and other fit indices leading to better adjustment and eventually, improved performance and retention. Spouse and family adjustment might fit as moderators in such a model. If no model Blackwell Publishers Ltd. 2001.
Meta-analysis of cross-cultural training 121
can be specified, researchers are at least well advised to control for as many of these factors as possible when evaluating the impact of their cross-cultural training programs. Acknowledgements We thank Robert Hausmann and Sumita M. Gupta for their help with data collection and coding.
Appendix A Stem-and-leaf-display for 16 effect sizes in adjustment data ⫺.9 I ⫺.8 I ⫺.7 I ⫺.6 I ⫺.5 I ⫺.4 I 2 ⫺.3 I ⫺.2 I 5 ⫺.1 I ⫺.0 I 2 +.0 I +.1 I 129 +.2 I 1133 +.3 I 03 +.4 I 35 +.5 I 77 +.6 I +.7 I +.8 I +.9 I Stem-and-leaf-display for 25 effect sizes in performance data ⫺.9 I ⫺.8 I ⫺.7 I ⫺.6 I ⫺.5 I ⫺.4 I ⫺.3 I ⫺.2 I ⫺.1 I ⫺.0 I 478 +.0 I 467 +.1 I 247 +.2 I 0345668 +.3 I 5 +.4 I 0399 +.5 I 38 +.6 I +.7 I 59 +.8 I +.9 I 122 International Journal of Training and Development
Blackwell Publishers Ltd. 2001.
Appendix B: annotated URL List for web resources on CCT Organisation
Web site
Comments
Cendant Intercultural http://www.cendantintercultural.com/ Partners with (formerly Bennett Andersen Consulting Associates) Meridian Global
http://www.meridianglobal.com/
Quintessential predeparture training for expatriates
Expat-Repat
http://www.expat-repat.com/
Non-profit, aimed at supporting expatriates and repatriates rather than B to B selling of training
Windham International
http://www.windhamint.com
Useful articles and data from surveys conducted with SHRM and the National Foreign Trade Council
Society for Human http://www.shrmglobal.org/ Resource Management maintains a site devoted to global HR issues
Many interesting links to conferences and country-specific sites
References *Asterisks denote articles that were included in the meta-analysis. Becker, L.L. (1982), ‘The Cultural Assimilator in Educational Settings: A Comparison of CrossCultural Training Techniques’, unpublished doctoral dissertation, University of NebraskaLincoln. *Befus, C. (1986), ‘A Treatment Approach for the Distress of Culture Shock Experienced by Sojourners’, unpublished doctoral dissertation, University of Miami. *Bhawuk, D.P.S. (1998), ‘The Role of Culture Theory in Cross-Cultural Training: A Multimethod Study of Culture-Specific, Culture-General, and Culture Theory-Based Assimilators’, Journal of Cross-Cultural Psychology, 29, 630–55. Birdseye, M.G. and Hill, J.S. (1995), ‘Individual, Organizational/Work and Environmental Influences on Expatriate Turnover Tendencies: An Empirical Study’, Journal of International Business Studies, 26, 787–813. *Black, J.S. (1988), ‘Workrole Transitions: A Study of American Expatriate Managers in Japan’, Journal of International Business Studies, 19, 277–94. Black, J.S. and Gregersen, H.B. (1991a), ‘The Other Half of the Picture: Antecedents of Spouse Cross-Cultural Adjustment’, Journal of International Business Studies, 22, 461–77. *Black, J.S. and Gregersen, H.B. (1991b), ‘Antecedents to Cross-cultural Adjustment for Expatriates in Pacific Rim Assignments’, Human Relations, 44, 497–515. Black, J.S. and Mendenhall, M. (1990), ‘Cross-Cultural Training Effectiveness: A Review and a Theoretical Framework for Future Research’, Academy of Management Review, 15, 113–36. Brislin, R.W. (1986), ‘Culture General Assimilator’, International Journal of Intercultural Relations, 10, 215–34. *Chemers, M.M. (1969), ‘Cross-cultural Training as a Means for Improving Situational Favorableness’, Human Relations, 22, 531–46. *Chemers, M., Fiedler, F., Lekhyanda D. and Stolurow, L. (1966), ‘Some Effects of Cultural Training on Leadership in Heteroculture Task Groups’, International Journal of Psychology, 1, 301–14. Blackwell Publishers Ltd. 2001.
Meta-analysis of cross-cultural training 123
*Christensen, C.P. (1984), ‘Effects of Cross-Cultural Training on Helper Response’, Counselor Education and Supervision, 23, 311–20. Church, A.T. (1982), ‘Sojourner Adjustment’, Psychological Bulletin, 91, 540–72. *Cui, G. and Awa, N.E. (1992), ‘Measuring Intercultural Effectiveness: An Integrative Approach’, International Journal of Intercultural Relations, 16, 311–28. Cushner, K. (1989), ‘Assessing the Impact of a Culture-General Assimilator’, International Journal of Intercultural Relations, 13, 125–46. Cuthill, S. (1997), Global Best in Class Study: Summary Report (Arthur Andersen and Bennett Associates, Andersen Worldwide, SC). Deshpande S.P. and Viswesvaran, C. (1992), ‘Is Cross-Cultural Training of Expatriate Managers Effective?: A Meta Analysis’, International Journal of Intercultural Relations, 16, 295–310. *Earley, P.C. (1987), ‘Intercultural Training for Managers: A Comparison of Documentary and Interpersonal Methods’, Academy of Management Journal, 30, 685–98. *Earley, P.C. (1994), ‘Self or Group? Cultural Effects of Training on Self-Efficacy and Performance’, Administrative Science Quarterly, 39, 89–117. Fiedler, F.E., Mitchell, T.R. and Triandis, H.C. (1971), ‘The Culture Assimilator: An Approach to Cross-Cultural Training’, Journal of Applied Psychology, 55, 95–102. *Gannon, M.J. and Poon, J.M.L. (1997), ‘Effects of Alternative Instructional Approaches on CrossCultural Training Outcomes’, International Journal of Intercultural Relations, 21, 429–46. Glass, G.V. (1976), ‘Primary, Secondary and Meta-Analysis of Research’, Educational Researcher, 10, 3–8. Goldstein, I.L. (1993), Training in Organizations: Needs Assessment, Development and Evaluation (3rd edn) (Pacific Grove: CA: Brooks/Cole). *Gudykunst, W.B., Hammer, M.R. and Wiseman, R.L. (1977), ‘An Analysis of an Integrated Approach to Cross-Cultural Training’, International Journal of Intercultural Relations, 1, 99–110. *Hammer, M.R. and Martin, J.N. (1992), ‘The Effects of Cross-Cultural Training on American Managers in a Japanese-American Joint Venture’, Journal of Applied Communication, 20, 161–82. Hannigan, T.P. (1990), ‘Traits, Attitudes and Skills that are Related to Intercultural Effectiveness and their Implications for Cross-Cultural Training’, International Journal of Intercultural Relations, 14, 89–111. *Harrison, J.K. (1992), ‘Individual and Combined Effects of Behavior Modeling and the Cultural Assimilator in Cross-Cultural Management Training’, Journal of Applied Psychology, 77, 952–62. Hedges, L.V. and Olkin, I. (1985), Statistical Methods for Meta-analysis (New York: Academic Press). *Hernandez, A.G. (1986), ‘Evaluating the Triad Model of Cross-Cultural Counselor Training’, Dissertation Abstracts International, 47, 2511. Hofstede, G.H. (1980), Culture’s Consequences: International Differences in Work-Related Values (Beverly Hills: CA: Sage). Hunter, J.E. and Schmidt, F.L. (1990), Methods of Meta-Analysis: Correcting Error and Bias in Research Findings (Newbury Park, CA: Sage). Hunter, J.E., Schmidt, F.L. and Jackson, G.B. (1982), Methods of Meta-Analysis (Newbury Park, CA: Sage). Kealey, D.J. and Protheroe, D.R. (1996), ‘The Effectiveness of Cross-Cultural Training for Expatriates: An Assessment of the Literature on the Issue’, International Journal of Intercultural Relations, 20, 141–65. Kenny, D.A. (1987), Statistics for the Social and Behavioral Sciences (Boston: Little Brown). Kenny, D.A. (1997), Meta-Software for Meta Analysis (computer software). Kirkpatrick, D.L. (1998), Evaluating Training Programs: The Four Levels (San Francisco: BerrettKoehler). Landis, D. and Brislin, R.W. (1983), Handbook of Intercultural Training (New York: Pergamon Press). *Landis, D., Day, H.R., McGrew, P.L., Thomas, J.A. and Miller, A.B. (1976), ‘Can a Black “Culture Assimilator” Increase Racial Understanding?’, Journal of Social Issues, 32, 169–83. *Lefley, H.P. (1985), ‘Impact of Cross-Cultural Training on Black and White Mental Health Professionals’, International Journal of Intercultural Relations, 9, 305–18. *Lefley, H.P. (1996), ‘Evaluating the Effects of Cross-Cultural Training: Some Research Results’, in H.P. Lefley and P.B. Pedersen (eds), Cross-Cultural Training for Mental Health Professionals, (Springfield. IL: C. Thomas) pp. 265–307. *Lightner, J.A. (1993), ‘Factors Leading to Cross-Cultural Adjustment and Satisfaction in Overseas Relocation’, unpublished Master’s thesis, Bowling Green State University, Ohio. *Malpass, R.S. and Salancik, G.R. (1977), ‘Linear and Branching Format in Culture Assimilator Training’, International Journal of Intercultural Relations, 1, 76–87.
124 International Journal of Training and Development
Blackwell Publishers Ltd. 2001.
*McDaniel, C., McDaniel N. and McDaniel A. (1988), ‘Transferability of Multicultural Education from Training to Practice’, International Journal of Intercultural Relations, 12, 19–33. *Nayar, E., Touzard, H. and Summers, D. (1968), ‘Training, Tasks and Mediator Orientation in Heterocultural Negotiations’, Human Relations, 21, 283–94. *Neimeyer, G.J., Fukumaya, R.P., Bingham, R.P., Hall, L.E. and Mussenden M.E. (1986), ‘Training Cross-Cultural Counselors: A Comparison of the Pro-Counselor and Anti-Counselor Triad Models’, Journal of Counseling and Development, 64, 437–9. *O’Brien, G.E., Fiedler, F.E. and Hewett, T. (1970), ‘The Effects of Programmed Culture Training upon the Performance of Volunteer Medical Teams in Central America’, Human Relations, 24, 209–31. Orwin, R.G. (1983), ‘A Fail-Safe n for Meta Analysis’, Journal of Educational Statistics, 8, 157–9. Parker, B. and McEvoy, G. (1993), ‘Initial Examination of a Model of Intercultural Adjustment’, International Journal of Intercultural Relations, 17, 355–79. Rahim, A. (1983), ‘A Model for Developing Key Expatriate Executives’, Personnel Journal, 62, 312–17. Rosenthal, R. (1980), New Directions for Methodology of Social and Behavioral Science: Quantitative Assessment of Research Domains (No. 5) (San Francisco: Jossey-Bass). Rosenthal, R. and Rubin, D. (1986), ‘Meta-Analytic Procedures for Combining Studies with Multiple Effect Sizes’, Psychological Bulletin, 99, 400–6. Schwarzer, R. (1996), Manual for Meta-Analysis Programs (Electronic document). Stajkovic, A.D. (1999), ‘Fitting Parametric Fixed Effect Categorical Models to Effect Sizes: A Neglected Meta-Analytic Approach in Organizational Studies’, Organizational Research Methods, 2, 88–102. *Steinkalk, E. and Taft, R. (1979), ‘The Effect of a Planned Intercultural Experience on the Attitudes and Behaviors of Participants’, International Journal of Intercultural Relations, 3, 187–97. Symonds, J., O’Brien, G., Vidmar, M. and Hornik, J. (1967), ‘Honduras Culture Assimilator’, unpublished article, University of Illinois, Urbana, IL. Thomas, K. (1996), ‘Psychological Privilege and Ethnocentrism as Barriers to Cross-Cultural Adjustment and Effective Intercultural Interactions’, Leadership Quarterly, 7, 215. Tolbert, A.S. and McLean, G.N. (1995), ‘Venezuelan Culture Assimilator for Training United States Professionals Conducting Business in Venezuela’, International Journal of Intercultural Relations, 19, 111–25. Tung, R. (1981), ‘Selection and Training of Personnel for Overseas Assignments’, Columbia Journal of World Business, 16, 68–78. Tung, R. (1982), ‘Selection and Training Procedures of U.S., European and Japanese Multinationals’, California Management Review, 25, 57–71. Tung, R. (1987), ‘Expatriate Assignments: Enhancing Success and Minimizing Failure’, Academy of Management Executive, 1, 117–25. United States Department of Commerce (1994), Survey of Current Business (Washington, DC: compiled by the Bureau of the Census, the Bureau of Foreign and Domestic Commerce, and the Bureau of Standards). Walton, S.J. (1990), ‘Stress Management Training for Overseas Effectiveness’, International Journal of Intercultural Relations, 14, 507–27. Wanous, J.P., Sullivan, S.E. and Malinak, J. (1989), ‘The Role of Judgement Calls in MetaAnalysis’, Journal of Applied Psychology, 74, 259–64. *Weldon, D., Carlston, D.C., Rissman, A,K., Slobodin, L. and Triandis, H.C. (1975), ‘A Laboratory Test of Effects of Culture Assimilator Training’, Journal of Personality and Social Psychology, 32, 300–10. Whitener, E.M. (1990), ‘Confusion of Confidence Intervals and Credibility Intervals in MetaAnalysis’, Journal of Applied Psychology, 75, 315–21. Windham International (1999), Global Relocation Trends 1999 Survey Report. Jointly prepared by the National Foreign Trade Council and the Institute for International Human Resources. http://www.windhamint.com/article/index.html. *Worchel, S. and Mitchell T.R. (1972), ‘An Evaluation of the Effectiveness of the Culture Assimilator in Thailand and Greece’, Journal of Applied Psychology, 56, 472–9. Yarbro, M. and Lynette, C. (1988), ‘An Assessment of the Ability of the Culture-General Assimilator to Create Sensitivity to Multiculturalism in an Educational Setting’, unpublished dissertation, University of Houston, Houston, TX.
Blackwell Publishers Ltd. 2001.
Meta-analysis of cross-cultural training 125