Corps de l’article

The best is the enemy of the good.

Voltaire

In his recent essay on government administered large-scale workplace surveys, John Godard (2001) has challenged researchers to consider the advantages and disadvantages of using these data sets in industrial relations research. Godard focused his critique on Australia’s Australian Work and Industrial Relations Survey (AWIRS), UK’s Workplace and Employee Relations Survey (WERS), and Canada’s Workplace and Employee Survey (WES). In what follows, we present a somewhat different viewpoint than does Godard on the role and future relevance of the WES to Canadian industrial relations research. We view the WES as good for extending IR related research, although it is clearly not the best. In contrast, Godard (2001: 27) concludes that: “The WES… could represent a ‘new dawn’ for researchers interested in various labour market and economic policy issues … But it may represent a ‘bad moon rising’ for mainstream Canadian IR research and possibly for the field of IR in general.”

We are of the view that it will likely do neither. While the WES will present significant new research possibilities, research on this data will neither transform labour economics nor labour policy. Similarly, while IR research using WES data stands to yield considerable insights into some important issues, it will not significantly alter the course or future of Canadian IR research; hence, neither will it be a “bad moon rising.” But we do consider the outlook for a “brighter day” to be excellent because the WES is an important source of data that augments the IR data that is currently available and will, consequently, expand the possibilities for rigorous empirical research in the IR field.

We begin by identifying the general areas and arguments put forward by Godard with which we are basically in agreement and concentrate on those with which we have some disagreement. Then, we proceed to focus more intensively on the WES itself. We present our case for the conclusion that IR research based on the WES will contribute to our knowledge base in IR and therefore contribute to a “brighter day.” We wish to emphasize at the outset that we agree with many of Godard’s observations. However, we tend to view progress in data gathering and, more generally, advances in the knowledge-building enterprise as unfolding more incrementally. The WES represents an imperfect but, nonetheless, very valuable new source of data that stands to advance IR research and, hence, the field of study.

Issues Associated with the General Case on Large Surveys

We agree with Godard (2001: 6–7) that the WES, as with other large scale government administered surveys, has a number of significant advantages including: excellent response rates, comprehensiveness, the ability to link employees with their employers and to follow them over (limited) periods of time, and a tendency to use more standardized measures. These represent substantial advantages relative to other sources of micro-level data.

Godard (2001: 8–9) also raises a number of general problems as well, including issues related to facilitating the construction of appropriate indices, the depth to which questions are investigated, the lack of research hypotheses underlying the survey construction and, more seriously, that the workplace constructs that the surveys are aimed at measuring are actually too complex to be captured by “single numbers” because they are “processes and relationships” and the surveys may have added little to our understanding of IR phenomena. While not taking absolute issue, there are several alternative considerations worth noting about each of these points.

Construction of Appropriate Indices. This point is one to which we are in considerable sympathy. The large government surveys yield data sets that lend themselves to types of statistical analysis (beyond essentially descriptive statistical methods) that tend to be most closely associated with a subset of the social science disciplines that contribute to IR research, notably economics. Designing the survey measures in a way that would broaden their usefulness across more disciplines would benefit IR and other social sciences as well.

Depth to Which Questions Are Investigated. While acknowledging the need for greater depth of detailed data on virtually all empirical phenomena, there are substantial tradeoffs to achieving greater depth. Greater complexity and detail in surveys does tend to lower response rates, even in government-administered surveys; highly detailed questions may apply to some respondents but not to others, so a degree of generality is required; and covering a broader set of issues or questions, instead of fewer but in greater detail, tends to broaden the scope of usefulness of the survey in terms of research questions as well as policy issues that can be addressed.

Lack of Research Hypotheses Underpinning Survey Construction. One of Godard’s (2001: 8) concerns is that, without a research approach in which hypotheses lead to survey design and content, the result can be “data driven” research agendas. One aspect of this point is that data availability can constrain the research questions that can be explored and, in some cases, the type of research analysis undertaken. (This is, of course, also the case across social science fields.) Using specific research hypotheses to inform data collection in government-sponsored survey projects may be a thornier issue than others. Hypotheses are typically either derived from structural models of behaviour (deductive research hypotheses) or are based on a body of (hopefully) fairly consistent empirical evidence (inductive research). In the former case, the challenge is that the particular theoretical model constructed to explain behavioural relationships could itself, by construction, affect the particular hypotheses that result from the theory.

IR researchers are constantly testing alternative models (and specifications of similar models), derived hypotheses, and so on. In IR, the variety of models used is compounded by the multi/inter-disciplinary nature of the field. The key advantage of the WES, as Godard (2001: 26) recognizes, is that it can be a valuable resource for examining workplace change and work systems. In this regard, it is important that the WES is appropriate in using either deductive or inductive approaches.

Real World IR Complexity. The complexity of IR processes and relationships and the interactions among the actors in IR makes studying them equally complex. It is not, however, at all obvious that survey designers have in mind an “…attempt to assign single numbers to these processes and relations” (Godard 2001: 9). In many cases, surveys are aimed at capturing observable conditions or outcomes and not processes or relationships. The main reason is that survey questions that yield empirical data typically better support effective hypothesis testing. Survey questions about processes or relationships often require the survey respondent to draw their own (personal) conclusions, while direct questions regarding perceived causality can result in responses that reflect a preconceived behavioural model used in crafting the question. It is often argued that case studies, for example, are therefore better approaches to studying these relational dynamics. Even so, a focus on outcomes alone in IR research can be a valid criticism since, in IR, as in other fields of enquiry, the processes can be as important as the outcome itself (e.g., political science, labour policy).

As with all research, empirical results do contribute to the “mapping” of at least some aspects of IR (Godard 2001: 9). By conducting studies of different aspects of the processes, relationships and outcomes of IR, using a variety of methodological approaches, and replicating studies over time, the mapping is made even more complete. It is entirely appropriate that the surveys enhance (even on the margin) our knowledge of IR; it is up to researchers, however, to develop the theory, and test the evidence, that improves our understanding. Large surveys are one, albeit imperfect, input into this process of theory building and testing.

The Canadian Workplace and Employee Survey as a Wolf in Sheeps Clothing

Godard’s (2001: 27) main conclusion regarding WES is that “it may represent a ‘bad moon rising’ for mainstream Canadian IR research.” This characterization stems from the conclusion that it has inherent flaws as an IR survey and that it may even be detrimental to the field of IR.

The WES as an IR Survey

The overall criticism of WES, as an IR survey, is that it has “weak industrial relations content” (Godard 2001: 26). Instead, Godard (2001: 25) characterizes WES as “basically a labour market and productivity survey” and he gives it a high grade in this regard. This characterization is an arbitrarily narrow view of its purpose and possibilities, in part because he employs a rather narrow definition of the field of IR (see Godard 2001: 5).

WES was not simply intended primarily as an industrial relations survey. What was it intended to do? One of the major challenges facing Canadian researchers wishing to examine labour market and workplace issues is that large-scale Canadian data bases have tended to focus on individuals (e.g., in the general population or the workforce). WES was explicitly aimed at filling the gaps in available firm-level, or “demand-side,” data, including information on establishments and workplaces (e.g., production systems, technology, and work practices), employment, and firm outcomes. These are critical aspects of labour economics but are, also, important aspects of industrial relations. While using WES to understand IR processes and relationships is beyond its scope, nonetheless, it stands to offer IR researchers some of the best, most comprehensive data on the workplace and on a range of important IR outcomes than were previously available in Canada. A comprehensive general IR survey would be of immense benefit, and the time has probably arrived for the IR community to consider advancing this objective.

The WES as Detrimental to IR

With regard to the potential for the WES to have a detrimental effect on the IR field, Godard (2001: 27–28) takes the view that it may end up potentially: (i) distorting research resources away from important IR issues; (ii) supporting the shift toward management and economic issues in the broader IR field; and (iii) as a result of the first two, compromising IR research itself. Further, Godard (2001: 25) surmises that “The survey seems to have been driven by the essentially managerialist policy paradigm that has become predominant both within the federal government (especially HRDC) and increasingly in applied economics or ‘policy studies’.”

One of the challenges that industrial relations scholars have carried forward over the years has been to define the parameters of the field. The difficulty in defining even approximate boundaries is made more difficult by the lack of a formal (unifying) theory in the field and the complexity of the issues studied which has, in turn, led researchers to explicitly embrace industrial relations as a multi- and inter-disciplinary field. Admittedly, we would define industrial relations somewhat more broadly, perhaps, than Godard does; and he acknowledges that his is a somewhat restricted definition. Thus we consider aspects of labour markets, and particularly institutional arrangements that interface with the labour market, to be very important, including the determination of pay policies, workplace practices, employment arrangements, etc. If managerialism is in the ascendancy, then we need to better understand that too. Rather than skewing research towards new areas, and detracting from others, as in a zero-sum game, we are of the view that the WES will extend the IR research agenda, attract new research support, and support research efforts that assist us in better understanding workplace change and emerging institutional arrangements.

We agree that the tide in IR practice, policy, and research has been toward what Godard refers to as the “managerialist” paradigm, although we view this as a manifestation of the more encompassing shift toward the ascendancy of economic markets and human resource management. This shift has been fostered by economic globalization and, in the case of Canada, by the FTA and NAFTA. The net effect has been to shift power in work relations in favour of management and, therefore, it is not surprising that unions have declined and management tends to drive workplace change. Putting aside the contentious issues of the ways in which this shift is detrimental, and to whom, we do perceive governments generally as supporting freer markets and shying away from regulation and labour market intervention. While some governments may, increasingly, disfavour unions as the “institutional check” of choice on growing managerial power, it does not necessarily follow that policy makers believe that no checks are required—they may favour a quite different model for accomplishing that outcome, or invent new ones, and IR scholars ought to study them as well.

In our estimation it is unlikely that WES was motivated by policy support for a particular workplace paradigm. This is not inconsistent with recognizing that the managerialist paradigm has become much more important and seeking to better understand associated workplace changes, which could be related to the rise of the managerialist paradigm or other forces such as technology. The sponsorship of HRDC of WES was more directly related to filling the critical data gaps noted above than to further any particular policy perspective.

Comprehensiveness and Access to the WES

We agree with Godard that it would have been beneficial to have had more extensive academic input into the survey—although IR scholars were directly represented. In hindsight, the marginal cost of including a few extra questions that may shed light on other important policy issues or basic research questions could have been low although, at the time of survey development, scaling back the scope of the survey was a concern. But the marginal cost of some updating may still be low, so that a mechanism to update and include a few additional questions, even now, would be useful.

Concerning access to the data, Godard (2001: 27) notes “independent researchers will have no direct access to the data. Statistics Canada will instead provide a “dummy” data set with which researchers can establish how they would analyse the data if they could. Researchers can then ask Statistics Canada to run the data accordingly, on their behalf.” This is currently the case for the employer-side component; the employee-side survey data is available to researchers through data centres. While we do not agree that this approach necessarily compromises the independence (unbiasedness) of the research (Godard 2001: 28), we do agree that remote access can be logistically constraining. To our knowledge, Statistics Canada places no conditions on the subject matter of research and is providing full support for remote access to the (employer side) WES data. The outstanding questions relate to: the adequacy of the number and geographical location of data centres; and whether requests for empirical analyses submitted through remote access will be truncated because of concerns over data confidentiality. It will be appropriate for all researchers to consider carefully whether logistically constrained access actually translates into bias.[1]

The Verdict on the Canadian Workplace and Employee Survey

We, too, bemoan the lack of comprehensive surveys that provide constructs that relate to relationships and processes in industrial relations. Certainly, WES does not fill this void. WES is not, however, a general IR survey, nor has it been advertised as such. We believe that most of the prospects for advancing IR research using WES derive from its value in deductive and quantitative (e.g., multivariate) research methods. WES is probably most relevant to the analysis of outcomes, not processes and relationships in IR—but this too is a critical aspect of IR research. WES will likely most closely relate to the field of labour economics (a field that is central to the study of IR). We probably most directly differ from the perspective of Godard in that we are persuaded that, in the long run, these characteristics of WES will allow it to substantially augment other methodological approaches and disciplinary perspectives in investigating IR issues—not detract from them.

The WES, Quantitative Analysis, and the Prospects for IR Research

Our main concern with WES and the successors to WES is with IR related research. By offering a much more comprehensive first-order mapping of workplace characteristics and practices, the descriptive data yielded by the WES is, in and of itself, of considerable value.[2] Picot and Wannell (1996: 14–16) provide a detailed discussion of the possibilities for human resources and labour relations research using the WES. Descriptive summary results from the 1999 WES data also suggest a range of industrial relations issues, for which data has previously been unavailable, that can now begin to be explored in areas related to organizational practices and workplace characteristics (e.g., technological change; organizational change), employment and work arrangements (e.g., nonstandard work), and job characteristics (e.g., technology use; participation).[3] Important issues such as the degree to which unions affect wage or earnings inequality, or performance, or within-firm inequality can now be explored empirically controlling for important factors that we were previously unable to account for (Picot and Wannell 1996: 14).

It is important to highlight the significant advantages that WES can of- fer precisely because it offers linked (matched) employer-employee data and because, over several years, WES will be available as a panel. Most observers of WES allude to these characteristics as strengths of the survey, including Godard, but these advantages are rarely elaborated upon. Linked employer-employee data allows the researcher to identify both individuals and employing firms. When the WES panel becomes available, not only will individuals and employing firms be identifiable, but also researchers will be able to follow individuals over a two-year period and follow firms over a six-year period. This will allow researchers to control for person- and firm-specific effects in ways that are not possible with standard data sets such as the Labour Market Activity Survey (LMAS), where the unit of analysis is the individual but employing firm information is generally lacking, or establishment surveys in which individual level information is generally lacking.[4]

For example, when empirically examining the determination of workers’ pay, individual effects (e.g., person-specific intercepts and coefficients on non-time-varying personal characteristics) are commonly recognized as important controls. However, equally important are firm effects (e.g., firm-specific intercepts and coefficients on firm-specific time-varying individual characteristics such as seniority). With WES, researchers will be able to simultaneously control for both individual and firm effects. Controlling for firm heterogeneity is also important when examining worker employment mobility. Indeed, for many research questions, a potentially important control is the interaction of worker and firm effects. For example, we can think of examining firm outcomes (e.g., productivity, technology adoption, or work organization) having controlled for worker characteristics that matter in determining these outcomes. Heretofore, this has been very difficult—if not impossible—to accomplish in Canada.[5]

Specific examples of the benefits of matched data, from recent research, illustrate the relevance of the WES to Canadian IR research.[6] Cahuc and Kramarz (1997) find empirical support for a model where authority (power) is delegated by the firm to workers in exchange for (lower) worker turnover (loyalty). Essentially, they examine the tradeoff between “voice” and “exit.” To accomplish this they use French employer-employee matched data that allows them to control for the effect of collective agreements on seniority. Bayard et al. (1999) use U.S. matched employer-employee data to decompose the overall sex wage gap into a sex segregation component and a sex wage difference component. They find a substantial portion of the sex wage gap is accounted for by the segregation of women into low paying industries, occupations, firms, and occupations within firms. Similarly, Gupta and Rothstein (2001) used Danish matched employee-employer data to examine the wage impact of firm level sex segregation. In both cases, matched data is necessary to examine the effects of within-firm sex segregation on wages.

These analyses illustrate the important ways in which the WES can support next-generation empirical research that explores a range of key issues in the domain of IR. This will serve to further deductive analyses in IR that will, in turn, strengthen the field. Kaufman (1993: 188) has argued that the science-building dimension of IR enquiry has been in the ascendancy and the problem-solving in relative decline—hence the decline in qualitative analyses. While this is quite consistent with the view that there has been a decline in inductive research in IR and a rise in deductive and quantitative analysis (see Whitfield and Strauss 2000), we are not of the view that this shift in scientific method is as much related to the state of IR as a field as it is to broader methodological trends in the social sciences.

Industrial relations has, of course, had a long and quite successful tradition of inductive research based, in part, on the availability of survey data but also, typically, on case studies, private surveys, interview research, and so forth (Kaufman 1993). Whether or not it is in decline in Canada is probably a debate worth bringing forward. Yet these methods continue to be used by researchers to successfully address very specific hypotheses, as in the case of recent empirical research on workplace practices and innovation, or change in union-management relations, that assist in both scientific enquiry and policy making (e.g., Osterman 1994, 2000; Betcherman et al. 1994; Verma and Chaykowski 1999). We view the deductive and inductive approaches to IR research as complementary; that is, wherever possible, it is probably preferable to combine descriptive, qualitative and multivariate empirical methods in studying complex phenomena in IR.