The effect of lag-structure identification when testing for fit

Organization Studies, Mid-Winter, 1994 by James J. Hoffman, Nancy M. Carter, John B. Cullen

Introduction

The concept of 'fit' is a central thrust for middle-range theories in many management disciplines. In this context, 'fit' refers to how variables, both internal and external to the firm, combine or match together to affect organizational performance. Although several recent studies have examined how variables, both internal and external to the firm, combine to affect organizational performance, one unexamined issue in this research involves selecting research designs that will capture the dynamic nature of the fit between these variables. Previous studies that have tested for fit have primarily relied on cross-sectional research designs where lag-structures have not been examined (Dewar and Werbel 1979; Argote 1982; Fry and Slocum 1984; Alexander and Randolph 1985). It has been argued that the dynamic co-alignment of internal and external variables necessitates the adoption of dynamic operationalizations of fit (Donaldson 1987; Venkatraman 1989). This requirement is most obvious when performance is being predicted. Rarely is a firm's performance instantaneously affected by changes in any single variable or combination of variables. Instead, the full effects of co-alignment on performance are not likely to be felt for several years. In such situations, static research designs fail to detect the effect of fit on performance.

One possible way to capture the dynamic co-alignment in fit relationships is to identify proper lag-structures. In this context, lag-structures are considered to be the amount of time between when the fit relationship is measured and when performance is affected. The issue of lag-structures is important, since one consequence of not taking into consideration the dynamic nature of fit relationships is the possibility that fit relationships will be missed during the testing process.

The purpose of the current study is to determine if the identification of proper lag-structures associated with the fit relationships being tested improves the effectiveness of the research design. This issue has been previously addressed in the organization theory literature but not in the context of the fit concept (see, e.g., Cullen et al. 1986). In order to carry out our objective, fit is tested using several lag-structure models.

Issues in Research Design Selection

The Use of Cross-sectional Research Designs when Testing for Fit

The primary research design that has been used to test for fit has been the cross-sectional research design. Usually, when a cross-sectional research design is used to test for fit, several organizations of different types are examined during one time period.

The literature indicates that cross-sectional research has many shortcomings associated with it. Kimberly (1976: 321-347) stated that interpreting cross-sectional results within organization processes necessarily assumes 'that all organizations go through identical, or at least similar, processes of structural evolution and that environmental and/or contextual influences are either constant or random'. He went on to say that, given existing evidence and theory, such an assumption about organizations would be clearly questionable.

Meyer (1972) also argues this point by stating that causal inferences drawn from data that describe large numbers of organizations at one point in time are risky. He lists two reasons for this assertion. First, causal inferences require assumptions about the time ordering of variables which are necessarily arbitrary when characteristics of organizations are considered. Second, the cross-sectional data on organizations are affected by boundary conditions which may artificially inflate or otherwise distort correlations.

One argument for the use of cross-sectional research designs is that companies willing to participate in longitudinal studies may be non-representative of the general population of organizations. Many firms willing to participate during the first wave of data collection may be unable to participate in the second round, due to merger, bankruptcy, loss of interest, etc. (e.g. Meyer 1972). Thus, it can be contended that cross-sectional research designs are less prone to selection bias than longitudinal designs. An argument can also be made that, although there are several drawbacks associated with cross-sectional research, these drawbacks are not problems when the variables under study are static. In this context, static variables are considered to be those organizational or environmental variables that do not change in their magnitude or direction over time. For example, if the same technology had been used by an organization during its entire existence, then technology would be a static variable for that organization. Dynamic variables are considered to be any organizational or environmental variable that changes over time. For most organizations performance would be a dynamic variable since it usually varies from year to year. In the case of the newspaper industry where the type of technology used to produce a newspaper has changed over the years, technology would be considered. a dynamic variable. It can be reasoned that in cases where variables are static, dynamic research designs are inappropriate since there would be no change over time in the variables to measure.