A Constrained Finite Mixture Structural Equation ... - CiteSeerX

A Constrained Finite Mixture Structural Equation Methodology for Empirically Deriving Strategic Types by Wayne S. DeSarbo Pennsylvania State University C. Anthony Di Benedetto Temple University Kamel Jedidi Columbia University Michael Song University of Washington

ISBM Report 6-2003

Institute for the Study of Business Markets The Pennsylvania State University 402 Business Administration Building University Park, PA 16802-3004 (814) 863-2782 or (814) 863-0413 Fax www.isbm.org [email protected]

A CONSTRAINED FINITE MIXTURE STRUCTURAL EQUATION METHODOLOGY FOR EMPIRICALLY DERIVING STRATEGIC TYPES

Wayne S. DeSarbo C. Anthony Di Benedetto Kamel Jedidi Michael Song*

March, 2003 Revised: December, 2003

* Wayne S. DeSarbo is the Smeal Distinguished Research Professor of Marketing at the Smeal College of Business Administration of Pennsylvania State University, University Park, Pennsylvania 16802. C. Anthony Di Benedetto is Professor of Marketing at the Fox School of Business Administration of Temple University, Philadelphia, Pennsylvania 19122. Kamel Jedidi is Professor of Marketing at the Graduate School of Business Administration of Columbia University, New York, NY 10027. Michael Song is the Michael L. & Myrna Darland Endowed Distinguished Chair in Entrepreneurship Professor of Marketing and Innovation Management at the University of Washington, Seattle, Washington 98195.

A CONSTRAINED FINITE MIXTURE STRUCTURAL EQUATION METHODOLOGY FOR EMPIRICALLY DERIVING STRATEGIC TYPES

ABSTRACT

Miles and Snow (1978) originally envisioned strategy as an agglomeration of decisions by which a firm group or strategic business unit (SBU) aligns itself with its environment. Accordingly, these authors conceptualized a classification of “strategic types” based on SBUs’ patterns of decisions into the now familiar Prospector-AnalyzerDefender-Reactor (P-A-D-R) framework. The Miles and Snow framework has been criticized by Hambrick (1983,1984) and others for ignoring industry and environmental factors, for lacking explanatory or predictive power in specific industries, and for being somewhat naively determined methodologically speaking. In this light, we devise a constrained latent structure structural equation based methodology to empirically estimate an “optimal” typology using data obtained from 216 SBUs in the U.S. as an illustration. This procedure allows for more comprehensive modeling and grouping of entities and simultaneously provides a diagnosis of the sources of heterogeneity producing the derived strategic groups. In addition, environmental and industry specific factors can be easily incorporated. Results show that our derived four mixed-type solution dominates the four-group P-A-D-R Miles and Snow classification (as well as a number of other nested model solutions) in terms of objective statistical fit criteria for this dataset, suggesting a more contingency-driven strategic stance adopted by these SBUs.

Keywords: Finite Mixtures, Strategic Types, Latent Structure Analysis, Firm Capabilities

2

1. INTRODUCTION The Miles and Snow (1978) strategic typology has been widely applied in both the management and marketing strategy literatures since its initial description in the late 1970s (Conant, Mokwa, and Varadarajan, 1990; Hambrick, 1983; McDaniel and Kolari, 1987; McKee, Varadarajan, and Pride, 1989; Ruekert and Walker, 1987; Shortell and Zajac, 1990). This generic typology classified businesses, based on their patterns of strategic decisions, into four categories: Prospectors, Defenders, Analyzers, and Reactors. Prospectors compete by constantly seeking out and developing new markets and products; Defenders focus on maintaining a secure niche in stable market segments; Analyzers sometimes defend existing markets, but occasionally enter a new product market with a “second-but-better” strategy; and, Reactors typically lack a consistent strategy. It is reasonable to expect that for different SBUs, different capabilities will have greater impacts on performance. In their strategic typology, Miles and Snow (1978) found a relationship between strategic type and firm capabilities. For example, Prospectors tend to use their skills at product engineering and R&D, and at anticipating new product or marketplace opportunities to compete (McDaniel and Kolari, 1987; Conant et al., 1990; Walker et al., 2003). Defenders compete by maintaining a protected niche in a stable product area by offering higher quality, superior service, or lower prices (Hambrick, 1983). They rely on their skills in marketing and market sensing, and in resource efficiency (Conant et al., 1990; Walker et al., 2003). Analyzers occupy a middle ground, exhibiting characteristics of both Prospectors and Defenders; the difficulty of maintaining an Analyzer position is that the organization must be relatively strong in capabilities associated with both Prospectors and Defenders. Miles and Snow (1978) also found that the first three "archetypal" strategic types all outperform the Reactors. The marketing and strategy literature defines capabilities as bundles of skills and knowledge, allowing SBUs to make best use of the assets they possess and to efficiently coordinate their activities (Day, 1990, p. 38). SBU-specific resources include both assets and capabilities; capabilities in particular are deeply rooted in the SBU's routines and practices and therefore are usually difficult for competitors to imitate (Dierckx and Cool,

3

1989). These SBU-specific resources are the SBU's main source of long-term competitive advantage and performance. It is up to SBU management to exploit these resources to their greatest advantage (Davidow and Malone, 1992; Mahoney and Pandian, 1992; Barney, 1991; Wernerfelt, 1984; Day, 1994). Several capabilities have been identified in the most recent literature in marketing and strategic management as being particularly critical to improved performance and success. These include technology, market linking, marketing, information technology, and management capabilities (Day, 1990, 1994; Day and Wensley, 1988). Technical capabilities allow the SBU to improve production process efficiencies, reduce costs, and increase competitiveness. Market linking capabilities such as market sensing and distribution channel linking let the SBU exploit marketplace opportunities. Capabilities in marketing, management, and information technology are also related to increased performance (e.g., profitability), and the SBU's ability to sustain competitive advantage (Conant et al., 1990; Jaworski and Kohli, 1993; Day, 1994). More recent work has investigated the interrelationships among SBU capabilities, environmental factors, and strategic type (DeSarbo, Di Benedetto, Song and Sinha 2003). Nevertheless, the original Miles-Snow research was limited in the number of industries and the range of capabilities studied. They did not systematically study all the possible linkages between capabilities and strategic type, nor did they attempt to prove the validity of their typology across other industry types. Similarly, they did not determine empirically whether, for example, Prospectors might outperform Defenders under some circumstances. Hambrick (1983) denoted a number of additional limitations of this typology: “The typology also has limitations. Its parsimony can be taken as an incomplete view of strategy. Its generic character ignores industry and environmental peculiarities. In fact, Miles & Snow (1978) and Snow and Hrebiniak (1980) stressed that the various strategic types would perform equally well in any industry, providing that the strategy was well implemented. This latter stance is inconsistent with the more typical view that an environment favors certain types of strategies…..The typology appears to warrant more development and testing. As noted, little consideration of the environment-strategy link has been given. No systematic evidence has been provided on how strategic types differ in their functional attributes…..” p.7

4

Hambrick (1983) attempts to relate differences in strategy to differences in performance conditioning on environmental and functional attributes. Many of his findings conflict with those predicted from Miles and Snow (1978), and suggests that deeper empirical research into the relationships between capabilities, strategic type, and performance across a wider range of industries is warranted. Later, Hambrick (1984) criticizes such general classification schemes: “Even though they are based on systematic empirical observation, they are not quantitatively based. Typologies represent a theorist’s attempt to make sense out of non-quantified observations. They may have the advantage of being “poetic” (Miles, 1983), that is ring true, often sounding very plausible. However, since they are the product of rather personal insight, they may not accurately reflect reality. Or, more likely, they may serve well for descriptive purposes but have limited explanatory or predictive power…..” p. 28 The research objective of this manuscript is to provide a quantitative methodology to more appropriately derive strategic typologies empirically in an attempt to resolve some of the criticisms leveled against the Miles-Snow typology over the years. We gather data on 216 SBUs/divisions located in the United States, representing selected industries. We use profitability and ROI as the measures of performance of the SBUs, and devise a constrained latent structure structural equation methodology to empirically derive an “optimal” typology for a given sample of data. Our procedure permits a more comprehensive modeling and grouping of SBUs, and provides a diagnosis of the sources of heterogeneity producing the derived strategic types. Our goal is not to uncover generic strategic types that could be necessarily generalized across all time periods, industries, data samples, etc. as this would be impossible to do. Rather, we propose a quantitative methodology to be utilized across any scenario in order to derive strategic types for a given empirical application (e.g., for a given time period, industry(s), etc.). We find that the four mixed-type solution derived by our proposed methodology for this particular application empirically dominates the Miles-Snow classification (as well as a number of other potential classifications) in terms of statistical criteria. While some similarity to the Miles-Snow typology is witnessed, we do, however, uncover differences in terms of capabilities and performance, that would have been missed if only the Miles-Snow typology had been applied to this sample.

5

2. THEORETICAL BACKGROUND 2.1. SBU Capabilities As noted above, capabilities can be defined as "complex bundles of skills and accumulated knowledge that enable SBUs to coordinate activities and make use of their assets" (Day, 1990, p. 38), and to create economic value and sustain competitive advantage. While the list of possible capabilities a SBU may have is enormous, in this study we focus our efforts on five categories of SBU capabilities that, according to the marketing and management strategy literature, are closely linked to sustainable advantage and long-term success (Day, 1990, 1994; Day and Wensley, 1988; Conant et al., 1990; Jaworski and Kohli, 1993). The specific ways each of these capabilities improves the performance of the SBU are briefly described below. Technology capabilities such as technology development, product development, production process, manufacturing process, technological change forecasting, and logistics allow an SBU to keep its costs down and/or to differentiate its offerings from those of competitors. These capabilities are activated by external challenges and opportunities presented by the market, the competition, and the environment. More efficient production effectively reduces cost, improves delivery consistency, and increases competitiveness and performance (Day 1994). Market linking capabilities include market sensing, channel and customer linking, and technology monitoring. These capabilities allow the SBU to compete more effectively by early detection of changes in the market environment. Market linking capabilities help the SBU bring key information in and thus allow it to be more responsive to changing customer needs. In addition, market linking capabilities allow the SBU to use technology capabilities effectively to exploit external possibilities (Day 1994). Marketing capabilities such as skill in segmentation, targeting, pricing and advertising allows the SBU to take advantage of its market linking and technology capabilities and to implement marketing programs more effectively. These capabilities also include knowledge of customers and competition as well as skill at integration of

6

marketing activities (Conant et al. 1990; McDaniel and Kolari 1987; De Sarbo et al. 2003). Information technology (IT) capabilities permit the SBU to diffuse technical and market information effectively throughout all relevant functional areas (Griffin and Hauser 1992; Narver and Slater 1990). Creative use of IT increases strategic flexibility and boosts the SBU's financial performance (Bharadwaj et al. 1999; Swanson 1994; Day 1994) and success with new products (e.g., Griffin and Hauser 1992, 1993; Song and Montoya-Weiss 2001). Management-related capabilities of all different types permit the SBU to take advantage of all of the above capabilities, and include human resource management, financial management, profit and revenue forecasting, and others (Walker et al., 2003; DeSarbo et al., 2003). 2.2. Strategic Types and SBU Capabilities Based on exploratory field studies conducted in textbook publishing, electronics, food processing, and health care, Miles and Snow (1978) developed a strategic typology (Prospectors, Analyzers, Defenders, and Reactors) classifying organizations according to enduring patterns in their strategic behavior. All three of these "archetypal" strategic types perform well, as long as the strategies are implemented effectively, and outperform the Reactors that do not show consistency in their strategic decisions. Miles and Snow (1978) found relationships between strategic types and SBU capabilities. Since Prospectors compete by using a first-to-market strategy, anticipating new product and market opportunities and through technological innovation, technological would be particularly important for them to succeed (Walker et al. 2003). Solid ties with the distribution channel, and good market research, are also important for Prospectors to ensure their R&D results in products that meet customer needs (Hambrick, 1983; McDaniel and Kolari, 1987; Shortell and Zajac, 1992). In addition, IT capabilities allow for better cross-functional integration critical to new product development (Robinson et al., 1992, Swanson, 1994; Moenaert and Souder, 1996; Griffin and Hauser, 1996; Bharadwaj, et al. 1999). By contrast, Defenders need to be able to offer higher quality and service, and/or lower prices, to succeed, and concentrate on

7

resource efficiency, cost-cutting, and process improvements. For them, marketing and market linking capabilities may be most critical for their continued success (Conant et al., 1990; Walker et al., 2003), since they rely on their more limited range of products or services, and protect their domains by offering higher quality, superior service, and lower prices (Hambrick, 1983). The objective of Miles and Snow (1978) was to find evidence of strategic types within the industries they studied. While they made some observations regarding business attributes (such as SBU capabilities), environmental factors (such as the extent of competition), and performance (e.g., Reactors are outperformed by the other three types), it was not their objective to quantify empirically the relationships among these variables. It does suggest, however, that deeper empirical testing of the relationships between an SBU’s environment and capabilities, its choice of strategy, and its resulting performance, is warranted. In some industries, SBUs may gravitate to strategic positions that are very like, or very unlike, those characterized by Miles and Snow, and empirically-derived strategic groups specific to that industry may be more appropriate than the Miles-Snow strategic typology. Further, given the various criticisms levied against this typology by Hambrick (1983, 1984), a more quantitative approach may prove beneficial. We propose a constrained latent structure structural equation procedure to accommodate many of the issues described above. In particular, our goal is to empirically derive strategic types from observed data and simultaneously obtain the relationships between SBU capability and profitability (as well as other factors such as environmental factors) per each derived strategic type. Model selection heuristics are developed which identify the appropriate number of strategic types. The model framework accommodates user specified constraints regarding the positivity of the estimated coefficients. Posterior probabilities of SBU membership in each derived strategic types are simultaneously estimated as well. We wish to provide a quantitative approach to the derivation of strategic types which optimizes an objective likelihood function, is based on relationships between strategy, environment, and performance, and can provide insight as to the nature and number of these derived strategic types and their SBU composition. Rather posit generic strategic types for all industries and applications, our goal is to develop a flexible,

8

adaptive methodology that can be employed to estimate strategic types in an empirical manner for any desired sample of SBUs. As to be shown, the quantitative framework provides a manner in which to compare the derived strategic types with those of Miles & Snow (1978) on objective statistical criteria related to explanatory power and prediction. 3. THE METHODOLOGY 3.1. Development of Firm Capability Scales Following the procedure of Churchill (1979), we used a multi-step instrument development procedure to develop and validate scales for the capabilities included in our study, as existing scales for these capabilities did not exist (except for marketing capability). In the first step, we identified relevant measurement scales from the marketing literature and grouped the scale items derived from these scales into the capability types. To this initial pool, we also added new items in instances where it was felt that not all the dimensions of the construct had been sufficiently covered. The scales were refined through in-depth focus interviews with managers in two SBUs (full details of the interview procedure are available from authors). According to these interviews, the managers perceived the scale items to be relevant and complete, and were able to rate their own SBU relative to major competitors easily on each scale item. In the second step, we assessed construct validity of the scales being developed by correcting ambiguous scale items or those that have possessed "different shades of meaning" to informants. Convergent and discriminant validity were established using Davis’s (1986, 1989) procedure. A team of seven judges (two professors and five doctoral students with background in measurement development) was asked to sort the items from the first step into the five capability scales, and inter-rater reliability was assessed (see Davis 1989). In the third step, we re-examined all scale items and eliminated inappropriate or ambiguous items or any that were inconsistently classified. The scales were then combined into an overall instrument for additional pretesting. The instrument was distributed to 32 managers in the two SBUs to further assess scale reliability and validity, and items that remained troublesome were deleted. The instrument was also distributed to

9

41 EMBA students taking a new product development class, and the results were factor analyzed (factor loadings and reliability test results are available from the authors on request). Two additional items that did not load appropriately were deleted, resulting in the final questionnaire. The five capability factors derived were as: Market Linking Capabilities. These include market sensing and linking outside the organization. The scale items used were derived from Day's (1994) descriptions of market linking capabilities. Using 0 to 10 scales (0 = "much worse than our competitors" and 10 = "much better than our competitors"), respondents were asked to rate their SBUs, relative to the top three competitors in their industry, on their capabilities in creating and managing durable customer relationships, creating durable supplier relationships with suppliers, retaining customers, and bonding with wholesalers and retailers. Technology Capabilities. These are capabilities relating to process efficiency, cost reduction, consistency in delivery, and competitiveness, and again the scale items were drawn from Day's (1994) descriptions of technological capabilities. Using the same 0 to 10 scales, respondents were asked to rate their SBUs relative to the three major competitors in their industry on their capabilities in new product development, manufacturing processes, technology development, technological change prediction, production facilities and quality control. Marketing Capabilities. In their study of marketing capability and strategic type, Conant et al. (1990) developed a set of scale items to measure marketing capabilities. The Conant et al. (1990) scale was used in this study. Respondents were asked to rate their SBU's knowledge of customers and competitors, integration of marketing activities, skills in segmentation and targeting, and effectiveness of pricing and advertising programs, relative to the top three competitors in their industry, on scales of 0 (much worse) to 10 (much better). Information Technology Capabilities. Based on the literature cited above (e.g., Narver and Slater 1990; Griffin and Hauser 1992; Day 1994; Bharadwaj et al. 1999), we developed a new scale of information technology capabilities, designed to measures the capabilities that help an SBU create technical and market knowledge and facilitate communication flow across functional areas. Using 0 to 10 scales as above, respondents were asked to rate the capabilities of their SBU's IT systems relative to the competition.

10

For example, respondents had to rate their IT systems on ability to facilitate technology and market knowledge creation, to facilitate cross-functional integration, and to support internal and external communication. Management Capabilities. Based on the relevant literature (e.g., Walker et al., 2003), we developed a set of six items measuring key management capabilities. Using 0 to 10 scales, respondents rated their SBUs, relative to their three major competitors, on their abilities to integrate logistics systems, control costs, manage financial and human resources, forecast revenues, and manage marketing planning. 2.3. Data Collection and Measures Our data were derived from a large-scale survey of 800 randomly selected U.S. companies listed in Ward’s Business Directory, Directory of Corporate Affiliations, and World Marketing Directory. There were four distinct phases of data collection: a presurvey, data collection on SBU strategies, data collection on relative capabilities, and phone/fax interviews for SBU information on profits and revenues. In the first stage, a one-page survey and an introductory letter was sent to selected firms requesting their participation and offering a set of research reports as an incentive to cooperate. Firms were asked to provide a contact person for a chosen, representative SBU/division. Of the 800 SBUs contacted, 392 agreed to participate and provided the necessary contacts at the SBU/division level. In the second stage, designated SBU managers were contacted by the researchers, and received a questionnaire and personalized letter. A three-wave mailing was used, as recommended by Dillman (1978). Data on the measures of strategic types were obtained from 308 SBUs in this phase. We used the 11-item strategic type scale previously developed and validated by Conant et al. (1990). In this phase, we also asked respondents to rate their confidence in their abilities to answer the questions thoroughly, and we eliminated individuals with low (below 6) levels of confidence from the final sample. In the third stage, a questionnaire including the relative capability scales was sent to the SBU managers, again followed up by a three-wave mailing. At the end of this stage, we had complete data on relative capabilities and strategic types from a total of 216

11

SBUs, which represents a 27.0% response rate. The following industries were represented: computer related products; electronics; electric equipment and household appliances; pharmaceuticals, drugs and medicines; machinery; telecommunications equipment; instruments and related products; air-conditioning; chemicals and related products; and transportation equipment. Annual sales of sample SBUs ranged from $11 $750 million, and SBU size ranged from 100 to 12,500 employees. We used the strategic type data collected in the second stage of the data collection process to classify the 216 SBUs/divisions responding to the third stage of data collection into the four Miles-Snow strategic types. We classified the SBU’s strategic type (Prospector, Analyzer, Defender, or Reactor) using the “majority-rule decision structure” of Conant et al. (1990), with one modification: for an SBU to be classified as a prospector or a defender, it must have at least seven “correct” answers out of the 11 items. Using this procedure, we classified the 216 SBUs/divisions as follows: 62 Prospectors, 79 Analyzers, 59 Defenders, and 16 Reactors. In the fourth and final stage, all 216 SBUs were contacted via phone or fax to obtain data on profits before taxes and revenues. One performance measure, profitability, is a measure of short-term profit, calculated as a percent profit margin by dividing SBU's current-year profit before tax by SBU's current-year revenue. The other dependent variable or performance measure was ROI, defined as the return on investments made by the SBU over the past three years. Our measure of ROI thus provides a longer-term view of profit performance within the SBU. 4. THE CONSTRAINED FINITE MIXTURE STRUCTURAL EQUATION METHODOLOGY 4.1.

The Model

We employ a structural equation approach to model the relationship between firm capabilities and performance. To capture firm heterogeneity, we use a finite-mixture approach which derives latent strategic types wherein firms are homogeneous in terms of their response parameters. Unlike previous multivariate finite mixture procedures (e.g.,

12

Jedidi, Jagpal and DeSarbo, 1997), our methodology allows the response parameters to be constrained (e.g., non-negative). This is an important feature in situations where theory imposes certain constraints on the parameters and/or the data suffer from multicolinearity problems. In the latter situations, it is well known that multicolinearity would lead to imprecise estimates as well as sign reversal. In addition, we permit “external analyses” and provide a host of nested constrained versions of the model where specific parameter subsets are fixed at designated values for hypothesis testing. Let i index firms (i=1,….,N) and s denote membership in a (a-priori unknown) strategic type (s=1,…..,S). Let η is be an (Mx1) vector of latent performance factors (e.g., profitability) for firm i in segment s and ξ is is a (Jx1) vector of latent firm capabilities (e.g., marketing, IT, management, etc.) and other exogenous unobserved factors. Conditional on membership in strategic type s, we postulate the following structural equation model:

ηis | s = β 0s + B s ξ is + ς is

(1)

where β 0s = ( β 01s , L, β 0sM )′ is an (Mx1) vector of intercept parameters, B s = (β1s , β 2s L , β sM )′

is an (MxJ) matrix of regression parameters that capture the effects of latent firm s capabilities ξ is on latent performance factors η is , β ms = ( β 1sm , L, β Jm )′ is a (Jx1) vector of

regression parameters for the mth equation, and ς is is an (Mx1) vector of error terms assumed to follow a multivariate normal distribution with null mean vector and covariance matrix Ψ s . 1 As firm capabilities are expected to impact performance positively, all the regression coefficients are constrained to be non-negative s (i.e., β jm ≥ 0; 1 ≤ j ≤ J ; 1 ≤ m ≤ M ). Assume that ξ is follows a multivariate normal

distribution with mean vector τ ξs and covariance matrix Φ ξs and is independent of ς is . Then The diagonal elements in Ψ s capture the residual variance of the latent performance factors in strategic group s after controlling for the effects of latent firm capabilities. The off-diagonal elements in Ψ s are covariance terms that capture the residual correlations between the performance factors due, for example, to omitted variables that affect all the endogenous variables.

1

13

the endogenous vector η is follows a multivariate normal distribution with (implied) mean vector

τηs = β 0s + B s τ ξs

(2)

and covariance matrix

Φηs = B s Φξs (B s )′ + Ψ s .

(3)

Let y i | s be a (px1) vector of observed performance measures (e.g., profitability, ROI, etc.) and x i | s is a (qx1) vector of observed firm capabilities measures (e.g., marketing, IT, management, etc.) for firm i in segment s . Then the latent exogenous and endogenous factors ξ is and η is are related to observed capability and performance measures by the following measurement models:

yi | s = Λsy ηis +εis ,

(4)

xi | s = Λsxξis +δis ,

(5)

where Λsy ( p × M ) and Λsx (q × J ) are factor loading matrices and ε is ( p × 1) and δ is (q × 1) are vectors of measurement errors in y i | s and x i | s respectively. Suppose that ε is and δ is follow independent multivariate normal distributions with zero mean vectors and diagonal covariance matrices Θ εs ( p × p ) and Θ δs (q × q) respectively. Then the joint y | s  (( p + q ) × 1) vector ∆ | s =  i  has a multivariate normal distribution with mean x i | s 

vector (see Jedidi, Jagpal, and DeSarbo 1997):

 Λsτ s  µ s =  sy ηs   Λ xτ ξ 

(6)

and covariance matrix : 14

Λ s Φ s (Λ s )′ + Θεs Λ sy B s Φξs (Λ sx )′  . Σ s =  ys ηs sy s s s s s   Λ x Φξ (B )′(Λ y )′ Λ x Φξ (Λ x )′ + Θδ 

(7)

If the strategic types are unobserved, then the unconditional distribution of the observed y  vector ∆ =  i  is a finite mixture of these S distributions. That is: x i  S

∆ i ~ ∑ ws f s (∆ i | µ s , Σ s ) ,

(8)

s =1

where w = ( w1 , L , wS )′ is the vector of the S mixing proportions such that ws > 0 and S

∑w

s

= 1, and f(▪) is the conditional multivariate normal density function MVN (µ s , Σ s ).

s

The likelihood function for a sample (∆1 ,L , ∆ N ) of i=1,…..,N randomly drawn observations from this finite mixture is then: N

S

LS = ∏∑ws fs (∆i | µs , Σs ),

(9)

i=1 s=1

where LS is a function of the strategic type-level parameters ws , B s , β 0s , Ψ s , Λ sy , Λ sx , Θ εs , Θ δs , and Φ ξs (s=1,…,S). The problem is to maximize LS (or, equivalently, log LS) with respect to these parameters given the sample data and the pre-specified number of strategic types S, while taking into account the constraints imposed on w and s β jm ≥ 0; 1 ≤ j ≤ J ; 1 ≤ m ≤ M above.

4.2.

Model Estimation

We use an E-M algorithm to maximize the likelihood function in Equation (9). For a detailed discussion of the E-M methodology, see Dempster, Laird, and Rubin (1977), Jedidi, Jagpal and DeSarbo (1997), Wedel and Kamakura (1999), and Wedel and DeSarbo (1994). To formulate this E-M algorithm, we begin by defining a latent indicator variable λis as:

15

1 iff firm i belongs to strategic type s,  0 otherwise.

λis = 

Assume that for a particular firm i , the unobserved vector λ i = (λi1 ,L , λiS )′ is i.i.d. multinomially distributed with probabilities w. That is, S

λ i | w ~ ∏ wsλis .

(10)

s =1

Then, the distribution of y i given λ i is:

∆i | λi

S

∑λ s =1

is

f s (∆ i | µ s , Σ s ) =

S

∏ [ f (∆ s

i =1

| µ s , Σ s )] is . λ

i

(11)

With L = ((λis )) considered as missing data and ∆ = ((∆ i )) as the input data matrix, the complete-data, log-likelihood function to be maximized is given by:

Log Lc (µ, Σ, w | ∆, L ) = −

N S 1 N S − ( J − 1) log( 2π )∑ ∑ λis − ∑ ∑ λis Σ s 2 2 i =1 s =1 i =1 s =1 N S 1 N S −1 ′ λ ( ∆ − µ ) Σ ( ∆ − µ ) + ∑∑ is i s s i s ∑ ∑ λis log( ws ), 2 i =1 s =1 i =1 s =1

(12)

where µ = (µ1 , L , µ S ) and Σ = ( Σ1 , L , Σ S ). This E-M algorithm described below maximizes (12) by iterating between two steps until convergence. The first is an E-step in which we compute the expected value of λ i given ∆ and provisional estimates for µ s , Σ s and w . The second is an M-step where we maximize (12) conditional on the newly estimated values L = ((λis )) to estimate all model parameters w s , B s , β 0s , Ψ s , Λ sy , Λ sx , Θ sy , Θ sx , and Φ ξs (s=1,…,S) using a non-linear optimization routine (i.e., a conjugate gradient procedure). We enforce the nonnegativity constraints of the response parameters by reparameterizing the constrained s = (γ sjm ) 2 . Hence our algorithm is general as it allows for both parameters as β jm

negativity2 and positivity constraints, and nests (unconstrained) finite-mixture structural 2

The negativity constraints can be enforced using the reparametrization β jm = − (γ s

16

s 2 jm

) .

equation models as a special case3. Once convergence is obtained, it is straightforward to obtain final estimates of the model parameters and the corresponding asymptotic covariance matrix. Specifically, we compute the asymptotic standard errors using the inverse of the empirical information matrix computed by taking the second-order derivatives of the unconditional likelihood function in equation (9) with respect to all model parameters. As we are interested in making inferences about the response parameters, the relevant second-order derivatives in the information matrix are taken with s and not γ sjm . 4 Note, within any iterate, we assign firms to each of the S respect to β jm

(latent) strategic types by using Bayes’ rule:

πˆ is =

(

ˆ wˆ s f s ∆ i | µˆ s , Σ s S

∑ wˆ g =1

g

(

)

ˆ f g ∆ i | µˆ g , Σ g

)

,

(13)

where πˆ is denotes the estimated posterior probability that firm i belongs to strategic type s. These probabilities represent a fuzzy classification of the N firms into the S strategic types. 4.3.

Nested Models

The finite mixture structural equation model described by equations (1), (4) and (5) is very general and subsumes a variety of models as special cases. If all variables are measured without error, the model reduces to a finite mixture, simultaneous equation model. If all variables are exogenous and error-free, the model reduces to a finite mixture of multivariate normal distributions (in this case, Λsx = I and Θ δs = 0 ). Reducing the model to either equation (4) or (5) produces a finite mixture confirmatory factor analysis model. In addition, our methodology allows the testing for invariance of certain parameters across strategic types. For example, we can make the measurement model parameters invariant across strategic types (i.e., Λsy = Λ y and Λsx = Λ x , s = 1, L, S ) ) if it is reasonable to assume that the groups react similarly to the measuring instrument for

3 4

The details of the estimation algorithm can be obtained from the authors. We thank an anonymous reviewer for drawing our attention to this point. 17

η s and ξ s . Likewise, we can impose invariance across groups on the covariance matrix of the exogenous factors (Φ ξs = Φ ξ ) and the covariance matrices of the error terms (i.e., Θ εs = Θ ε and Θ δs = Θ δ ) . We can impose several other model restrictions depending on the context being studied and the theory being tested. For example, if the endogenous and exogenous factors are all measured with single indicators, then we must set Λ sy = I, Λ sx = I, Θ εs = 0 and Θ δs = 0. In such ways, we can utilize likelihood ratio tests

(LRT) conditioned on S to examine the sources of heterogeneity producing the derived strategic groupings. For example, models can be estimated where the estimated exogenous and/or endogenous means are equal across derived strategic types. Or, models can be estimated where Bs = 0 for s=1,….,S indicating no relationship between the exogeneous and endogeneous variables across the strategic types. 4.4.

Model Selection

In general, the number of strategic types is not known a priori. To determine the number of derived strategic types (the value of S), the estimation procedure must be run for varying values of S. A number of model selection heuristics have been suggested for such finite mixture models. Bozdogan and Sclove (1984) discuss the use of Akaike’s (1974) information criterion (AIC) for choosing the number of strategic types in such finite mixture models. Accordingly, one would select S to minimize: AIC S = - 2 ln L S + 2N S

(14)

where N S is the number of free parameters. Bozdogan (1994) has proposed modifying this AIC heuristic where the penalty coefficient multiplying NS is 3 (instead of 2) for such finite mixture models. We shall use this refer to it as the “modified” AIC (i.e., MAIC). Recently, Andrews and Currim (2003) find that the MAIC outperforms AIC, BIC, CAIC, ICOMP, etc in terms of accurately identifying regression mixture model components across a wide variety of model specifications and data configurations. As such, we shall be using this criteria for designating the number of strategic groups in our particular model specification. 18

Note, we have programmed this methodology to also accommodate “external analyses” for comparative hypothesis testing and model comparisons. For example, one can test a derived estimated solution against any proposed alternative solution (e.g., the Miles and Snow four strategic types), and utilize one of the many information heuristics to designate which solution was “better” fit by the data. Here, for example, given an alternative, pre-specified taxonomy of firms, one can fix the posterior probabilities of membership, π is , average them to obtain estimates of the mixing proportions, and just use the M-Step of the EM algorithm to obtain estimates for B s , β 0s , Ψ s , Λ sy , Λ sx , Θ sy , Θ sx , and Φ ξs for each designated strategic type (s=1, …S). We will utilize this handy feature of the

proposed methodology to compare our derived solution with that suggested by Miles and Snow via their classic four strategic type solution. As mentioned earlier, this statistical comparison can be extended to any prior strategic type structure. Before proceeding, it is important to establish that this specified finite mixture structural equation model is identified. Jedidi, Jagpal, and DeSarbo (1997) have demonstrated that finite mixture of structural equation models are identified provided that the structural equation model is identified5 for known strategic types, and the data (within strategic type) follow multivariate normal distributions. Note, our model satisfies both conditions and is therefore identified. First, such multivariate regression type models are identified for known strategic types given non-discrete, non-singular data (cf. Hennig, 2000).6 Second, we explicitly make the parametric assumption that all the error terms follow a multivariate normal distribution.

5. THE EMPIRICAL RESULTS Figure 1 depicts the structural equation model that we specified to empirically derive the strategic typology of SBUs. Given the relatively small sample size of firms, 5

See Bollen (1989, pp. 88-103 and pp. 238-246) for a detailed discussion on the identification of structural equation models.

19

we utilize the first principal component scores derived within each set of capability items as exogenous variables in the subsequent analysis.7 We estimate the full model while varying the number of (latent) strategic types from 1 through 5. Because of the relatively small sample size, we set the covariance matrix of the exogenous factors to be invariant across groups (i.e., Φ ξs = Φ ξ , s = 1, L , S ). To avoid local optima problems, we have estimated each model at least twenty times using different random starting values and we selected the solutions with best log-likelihood values. ---INSERT FIGURE 1 ABOUT HERE--5.1. Aggregate S=1 Results

The overall relationship between firm capabilities and profitability/ROI was estimated for the full sample by our proposed procedure for S=1 (see Table 1). This analysis is equivalent to estimating an ordinary structural equation model as long as all the regression coefficients estimated remain non-negative. ---INSERT TABLE 1 ABOUT HERE--Table 1 shows that, overall, technology capabilities are most strongly related to performance. Four of the five capabilities (technology, IT, marketing, and market linking) are significantly related to profit performance (coefficients = 0.436, 0.285, 0.162 and 0.156 respectively; coefficients significant at the 0.05 level are indicated in bold), while only technology is significantly related to ROI performance (coefficient = 0.587). Thus, in the aggregate sample case, we find that the drivers of profit and ROI appear to

6

Hennig (2000) points out that the identification of finite mixture regression models is problematic when the regressors are discrete. We thank an anonymous reviewer for drawing our attention to this point. 7 With a small sample size of 216 firms, it is impossible to estimate all the parameters of the finite mixture structural equation model. In the context of our study, a fully specified structural equation model would require the estimation of 84 parameters per group excluding the mixing proportions (22 factor loadings, 27 error variances, five factor means, 15 exogenous factor variance and covariance elements, 3 structural variance-covariance elements, and 12 regression coefficients). The number of parameters would be still high even if we impose across-group invariance restrictions on certain parameters. As suggested by one anonymous reviewer, the limitation of using the first principal component of a separate factor analysis is that the errors in the estimated factor scores are ignored in the second-stage analysis. 20

be different: IT, market linking, and marketing are significantly related to short-term profit but not to longer-term ROI performance. 5.2. The Constrained Latent Structure Structural Equation Model Results

The aggregate sample S=1 solution presented above suggests that, overall, technology-related capabilities are most closely related to performance, and that the drivers of profit and ROI performance are somewhat different (at least for this sample of U.S.-based firms). These conclusions ignore the possibility of strategic types, that is, they leave un-addressed the issue of whether different capabilities are more critical to performance than others for different groups or types of firms. To investigate this, we analyze this set of firms using the constrained latent structure structural equation methodology described earlier which explicitly models the observed relationships between capabilities and performance, as well as the mean levels of each, and choose the "best" solution using minimum MAIC criteria. Table 2 presents a summary of the various goodness-of-fit heuristics for our proposed constrained latent structure structural equation methodology as applied to this data set. The analysis was performed in S=1,2,3,4, and 5 strategic types, with the MAIC heuristic designating S=4 derived strategic types as the “optimal” solution (minimum MAIC heuristics = 4090.20 at S=4). Note that we also force the firms into the MilesSnow strategic types according to the “external analysis” option described earlier. We compare the fit of the Miles-Snow solution to that obtained for the empirically-derived solution to assess any marginal improvement gained by using the empirically derived strategic typology for this sample. These results are also shown in Table 2. ---INSERT TABLE 2 ABOUT HERE--It is clear from Table 2 that all of the empirically-derived solutions dominate the Miles-Snow strategic typology in terms of this information heuristic (MAIC for the Miles-Snow solution =, 4287.00). Thus, while insights can be gained about the relationships between capabilities and performance by examining the results for the original Miles-Snow strategic typology, an even more accurate picture of the strategies

21

employed by these firms, and their resulting performance, can be obtained from the empirically-derived classification scheme. Table 2 also presents the comparison of several nested models. The full, unrestricted model result is presented in the top panel of Table 2 where S=4 is indicated by MAIC to be the “best” solution. The middle panel of Table 2 examines various nested models and sources of heterogeneity for the S=4 solution. The “Null Beta Coefficients” solution constrains Bs = 0 for all s=1,….,S indicating no relationship between firm capabilities and performance. As shown by the LRT vs. the full model, this nested model can be rejected. The “Fixed Endogenous Means” constrains the estimated means for the two performance variables to be equal across derived strategic types. That is, strategic types are formed that do not vary with respect to the way they perform. As shown in the middle panel of Table 2, this solution is also rejected according to the LRT vs. the full model. Finally, the “Fixed Exogeneous Means” solution constrains the estimated means for the firm capabilities components to be equal across derived strategic types. This solution is also rejected by the LRT vs. the full model. Thus, it appears that for this particular data set, one needs to explicitly consider heterogeneity with respect to both sets of variables (endogeneous and exogeneous), as well as the relationships between the two in order to derive appropriate strategic types here. All three components are necessary in the modeling effort (the Miles and Snow typology conceptually considers mainly the levels of firm capabilities). The lower panel of Table 2 displays the comparison the full model S=4 solution, the Miles and Snow typology, and respective multivariate mixtures for each, which we discussed as special cases in Section 4.3. Again, the MAIC heuristic [KAMEL: CAN WE USE AND REPORT THE LRT HERE AS WELL?] favors the derived full model solution. 5.3. External Analysis: The Miles-Snow Strategic Types

We first examine the external analysis results by strategic type using the original Miles-Snow typology (see Table 3). Table 3 shows that somewhat different relationships between capabilities and performance exist for each strategic type. For prospectors and analyzers, technology capabilities are most strongly related to both profit and ROI performance, while for defenders, the capabilities most significantly related to profit

22

performance are market linking and marketing (0.494 and 0.239 respectively). These findings are essentially in line with expectations from the Miles-Snow typology, though some findings were surprising. For example, both technology and IT capabilities were significantly related to ROI for defenders, but market linking was not (coefficients for technology and IT = 0.805 and 0.341 respectively). One might anticipate from the Miles-Snow typology that for analyzers, both technology and marketing skills should be significantly related to profit performance, but in fact only technology skills were found to be significant (0.512 and 0.182 respectively). No significant relationships were found between capabilities and profit or ROI performance for reactors. We conclude that the drivers of profit and ROI performance appear to be somewhat different across MilesSnow strategic types. ---INSERT TABLE 3 ABOUT HERE--The middle panel of Table 3 displays the estimated endogeneous and exogeneous means for the four Miles-Snow strategic types. As predicted by Miles and Snow, firms following any of the three archetypal strategic types were found to outperform reactors (raw profit performance of prospectors, analyzers, and defenders were 14.71, 7.34, and 5.81 respectively, while the average profit for reactors was –8.48). Interestingly, this expectation was not confirmed for ROI performance, which was not shown to differ significantly across the four types. (Note: since profitability is reported as a one-year profit figure, and ROI is reported over a three-year time period, it is possible for an SBU to show negative profit and positive ROI.) Examining the raw mean scores of each group on each capability scale item, prospectors were found to be strongest in technology and IT capabilities, while defenders were strongest in marketing and market linking capabilities. For most capability scale items, the differences in means across strategic type are significant (as indicated in Table 3). In almost every case, the mean score of analyzers was between the mean scores of prospectors and defenders. These findings are consistent with Miles and Snow. To summarize the estimated means here:

23

•

Prospectors have the highest standardized mean scores on technology and IT capabilities (0.240 and 0.645 respectively), and one of the highest standardized mean scores on management capabilities (0.119). Prospectors have low scores on marketing and market linking capabilities (-0.265 and – 0.346 respectively). (Note: the differences are only statistically significant in the cases of marketing and technology capability.)

•

Defenders are the polar opposites of prospectors. They have the highest standardized mean scores on marketing and market linking capabilities (0.501 and 0.255 respectively), and lower scores on technology, IT, and management (-0.207, -0.428, and -0.181 respectively).

•

Analyzers, as expected from the Miles-Snow model, have standardized mean scores midway between prospectors and defenders on each of the five capability scales.

•

Reactors score among the lowest on all capabilities except market linking and management (0.094 and 0.146 respectively).

•

Prospectors show the highest short-term profitability (0.563). The differences across the ROI standardized mean scores are not significant, but defenders have the highest standardized ROI (0.141).

5.4. The Derived "Mixed-Type" Strategic Group Solution

Next, we examine the constrained latent structure structural equation model estimates for the relative importances of firm capabilities by strategic type for the derived, "mixed-type" solution (see Table 4). We characterize our solution as "mixedtype" as each derived group or type, as to be shown shortly, is composed of a mix of Miles-Snow strategic types. ---INSERT TABLE 4 ABOUT HERE--Table 4 clearly shows a different set of significant relationships between firm capabilities and performance. For Strategic Type 1, IT has the greatest influence on profit performance (coeff. = 0.497), while technology and market linking capabilities

24

have a greater influence on ROI (coefficients = 0.759 and 0.257 respectively). Four of the five capabilities (marketing, technology, market linking, and management) significantly influence profit performance for Strategic Type 2 (coeffs. = 0.510, 0.418, 0.339, and 0.230 respectively, though none of the capabilities has a significant impact on ROI. For Strategic Type 3, the most significant influences on profit performance are technology, market linking, and management, though all five capabilities have significant influences on profit (coeffs. for technology, market linking, management, IT, and marketing are 0.484, 0.248, 0.228, 0.167, and 0.148 respectively), but only marketing significantly impacts on ROI (0.666). Finally, Strategic Type 4 shows yet another different pattern: technology and marketing influence profit performance (1.555 and 0.333), while IT influences ROI (0.447). Table 4 also includes the estimated means of exogenous and endogenous variables by strategic type, analogous to Table 4 for the Miles-Snow types. The analysis of exogenous variables provided additional insight for interpretation of the strategic types. Strategic Type 1 has the highest estimated standardized mean in technology capability (1.244), and is second-highest in market linking, IT, and management capabilities (0.131, 0.093, and 0.191 respectively). Strategic Type 2, conversely, is highest in marketing, market linking, and management capabilities (0.776, 0.320, and 0.441). Strategic Type 3 is highest in IT capability (0.109) but scores relatively low in all other capabilities. Strategic Type 4 is higher in market linking and management capability than Strategic Type 3 (-0.083 and –0.l19 respectively). Analysis of the endogenous variable means shows that: Strategic Type1 outperforms all other types on both profit and ROI, followed by Types 4 and 3 respectively, and Strategic Type 2 shows the poorest performance. Finally, we explore the correspondence between the Miles-Snow typology and the empirically-derived "mixed-type" typology via the cross-tabulation of firm membership presented in Table 5. As this Table shows, Strategic Type 1 (the highest-performing type) is comprised of approximately one-third prospectors and one-third analyzers. There are slightly fewer defenders, and a small number of reactors. Interestingly, Strategic Type 4 (the second-highest-performing type) has almost the same distribution of MilesSnow types: one-third each of prospectors and analyzers, with the remainder being

25

defenders and reactors. Strategic Types 2 and 3 (the lowest-performing types) are comprised of almost one-half analyzers, with proportionately less defenders and prospectors. ---INSERT TABLE 5 ABOUT HERE--Taking all the information presented in Tables 4 and 5 (as well as computing raw scale item means which we do not show to save space), we can finally develop rich descriptors of each of the four empirically-derived strategic types. Strategic Type 1: New Product/Market Seekers with Technology Strengths. This strategic type has a large number of prospector and analyzer firms, as well as several of the more aggressive defenders. Typical of prospectors, members of this strategic type have notable strengths in technology, in particular new product development. They also have strengths in information technology, particularly IT that supports new product development, cross-functional integration, and technology knowledge creation. Judging from the performance measures, Strategic Type 1 seems to be able to reap short-term profitability and also to translate new product success into superior longer-term ROI performance. Strategic Type 2: Defensive Firms with Marketing Strengths. Unique among the four strategic types, this type is made up largely of analyzers and defenders. Typical of defenders, this type tends to excel in the marketing capabilities, including information gathering, segmenting and targeting markets, pricing, and advertising. It also tends to be strong in management capabilities, including financial management, human resource planning, marketing planning, and logistics. Despite these capabilities, however, this strategic type is outperformed by other SBUs in terms of both short-term profit and longterm ROI. Strategic Type 3: Second-But-Better Firms with IT Strengths. This group seems to be most closely aligned with analyzers, who use their second-but-better strategies and capabilities in IT (in particular, IT to support new product development, cross-functional integration, and technology knowledge creation) to get a competitive edge. This strategy type is second only to Type 1 on all of these IT capabilities. This type’s profit

26

performance, however, is relatively low, surpassing only Strategic Type 2 in terms of both profit and ROI. Strategic Type 4: New Product/Market Seekers with Market Linking and Management Strengths. Like Strategic Type 1, this type is comprised mostly of analyzers and prospectors, with many defenders as well. Whereas Strategic Type 1 is strong in both IT and technology capabilities, Strategic Type 4 is noticeably weaker in technology capabilities, but is in the middle of the pack in terms of other capabilities. What distinguishes this strategic type from the poorer-performing Strategic Type 3 is relative strength in market linking and management capabilities. Strategic Type 4 is the secondbest performing type in terms of both short-term profit and longer-term ROI. 6. DISCUSSION The Miles and Snow (1978) strategic typology categorizes SBUs according to their patterns of strategic decisions by which they align themselves with their environment. In this study, we devised a constrained latent structure multivariate regression methodology and empirically derived a four-group mixed-type strategic typology. Our typology improved upon the Miles and Snow typology in terms of statistical fit, indicating that it better captures the interrelationships among the firm capability and performance variables we considered. Our findings suggest that strategic decision-making by SBUs is environmentally context-dependent: different capabilities lead to improved profit performance for different strategic types. Consistent with other recent work on strategic typology (see DeSarbo et al. 2003), we find that the Miles-Snow typology provides a useful means of strategic decisionmaking, but can be improved upon by considering contingencies such as specific firm capabilities and also by considering different measures of performance. Our mixed-type typology augments the Miles-Snow typology in that we found differences among the types that can be explained in terms of capabilities and performance, that might not have been evident had the classic Miles-Snow typology been used. Further, we also uncovered interesting patterns between firm capabilities and different performance measures. One constellation of firm capabilities (Strategic Type 1) was linked to highest performance in

27

both short-term profit and longer-term ROI; a second set of firm capabilities was linked to the second-highest performance level (Strategic Type 4), while the other strategic types were also-rans. Another insight concerned the two highest performers, Strategic Types 1 and 4, taken together. Both of these were primarily (though not exclusively) made up of firms that were classified as prospectors and analyzers using Miles-Snow; there were proportionally less defenders in each strategic type. The results suggest that there are two ways to success. The best-performing strategic type, Type 1, showed superior capabilities in technology and IT (which would be consistent with prospector strategy), but was not necessarily weak in any of the other capabilities. Strategic Type 4, by contrast, did not have a profile consistent with prospector strategy: these SBUs used fairly strong capabilities in IT, and good market linking capabilities, to offset weaknesses in technology and marketing capabilities and still be profitable. This insight was obscured when considering only the Miles-Snow solution. In sum, different combinations of firm capabilities seem to drive different measures of performance; and using the methodology described in this article permitted greater insights into the dynamic relationships among the capability and performance variables. There are certain limitations to our study. We are unable to confirm whether the strategic type solution we estimated is generalizable to other industries, countries, or geographical regions. An important benefit or advantage of our proposed methodology, however, is that one can derive an “optimal” strategic typology of firms specific to any specified industry, given data availability. Our underlying premise is that the Miles and Snow typology is too generic to be blindly applied to all empirical contexts, and ignores other important factors including firm performance, capabilities, environment, nd so forth. Extensions of this study could focus on identifying the specific strategic types found in other environmental contexts. Nevertheless, we believe the constrained latent structure multivariate regression methodology presented here can be successfully and flexibly used in understanding strategic decision-making and performance outcomes in a wide range of contexts and industries.

28

REFERENCES

Akaike, H. 1974. A new look at statistical model identification. IEEE Transactions on Automatic Control. 6, 716-723. Andrews, R.L. and I. S. Currim. 2003. Retention of latent segments in regression-based marketing models. International Journal of Research in Marketing. 20 315-321. Arminger, G., P. Stein, G. Witterberg. 1999. Finite mixtures of covariance structure models with regressors. Psychometrika 64, 475-494. Barney, Jay. 1991. Firm resources and sustained competitive advantage. Journal of Management 17, 99-120. Bharadwaj, Anandhi S., Sundar G. Bharadwaj, Benn R. Konsynski. 1999. Information technology effects on firm performance as measured by Tobin's Q. Management Science 45 (7), 1008-1024. Bollen, Kenneth A. (1989), Structural Equations with Latent Variables, New York: Wiley. Boyles, R. A. 1983. On the convergence of the EM algorithm. Journal of the Royal Statistical Society, Series B 45, 47-50. Bozdogan, H. 1987. Model selection and Akaike’s information criterion (AIC): The general theory and its analytical extensions. Psychometrika 52 345-370. Bozdogan, H., S. L. Sclove. 1984. Multi-sample cluster analysis using Akaike’s information criterion. Annals of the Institute of Statistical Mathematics36 163180. Bozdogan, H. (1994). Mixture Model Cluster Analysis Using Model Selection Criteria and the New Information Measure of Complexity. Proceedings of the First U.S.-Japan Conference on the Frontiers of Statistical Modeling, Kluwer, 69-113. Cohen, J. A. 1960. A coefficient of agreement for nominal scales. Educational and Psychological Measurement 20 37-46. Conant, Jeffrey S., Michael P. Mokwa, P. Rajan Varadarajan. 1990. Strategic types, distinctive marketing competencies and organizational performance: A multiplemeasures-based study. Strategic Management Journal 11 365-383. Davidow, W. H. and M. S. Malone1992. The Virtual Corporation. New York: Harper Business. Davis, Fred D. 1986. A technology acceptance model for empirically testing new end 29

user information systems: Theory and results," unpublished doctoral dissertation, Massachusetts Institute of Technology. Davis, Fred D. 1989. Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly 13(3) 319-339. Day, George S. 1990. Market Driven Strategy: Processes for Creating Value. New York: Free Press. Day, George S. 1994. "The Capabilities of Market-Driven Organizations," Journal of Marketing, 58(4) 37-52. Day, George S., Robin Wensley. 1988.Assessing advantage: A framework for diagnosing competitive superiority.Journal of Marketing, 52(2) 1-20. Dempster, A.P., N. M. Laird, D. B. Rubin (1977). Maximum likelihood from incomplete data via the EM algorithm. Journal of the Royal Statistical Society, Series B, 39 1-38. DeSarbo, W.S., E. A. Edwards (1996). Typologies of compulsive buying behavior: A constrained clusterwise regression approach. Journal of Consumer Psychology, 5 231262. DeSarbo, Wayne S., William Cron. 1988. A conditional mixture maximum likelihood methodology for clusterwise linear regression. Journal of Classification, 5 249-289. DeSarbo, Wayne S., C. Anthony Di Benedetto, Michael Song, Indrajit Sinha. 2003. Modeling a contingency framework of strategic choice involving strategic types, firm capabilities, and environmental uncertainty. working paper.. Dierckx, I., K. Cool. 1989. Asset stock accumulation and sustainability of competitive advantage. Management Science 35(12) 1504-1511. Dillman, D.A. 1978. Mail and Telephone Surveys: The Total Design Method. New York: Wiley. Dolan, C. V., J. J. L. van der Mass. 1998. Fitting multivariate normal mixtures subject to structural equation modeling. Psychometrika, 63, 227-253. Duda, R. O., and P. E. Hart. 1973. Pattern Recognition and Scene Analysis. New York: Wiley.. Dyer, Barbara. 1997. The impact of strategy on conflict: A cross-national comparative study of U.S. and Japanese firms. Journal of International Business Studies 28(3) 467493.

30

Griffin, Abbie and John R. Hauser. 1992. Patterns of communication among marketing, engineering and manufacturing: A comparison between two new product teams. Management Science 38 360-373. Griffin, Abbie and John R. Hauser. 1993. The voice of the customer. Marketing Science 12(1) 1-27. Griffin, Abbie and John R. Hauser. 1996. Integrating R&D and marketing: A review and analysis of the literature. Journal of Product Innovation Management 13(2) 191-215. Hambrick, Donald C. 1983a. Some tests of the effectiveness and functional attributes of Miles and Snow’s strategic types. Academy of Management Journal 26 (March) 5-25. Hambrick, Donald C. 1983b. An empirical typology of mature industrial-product environments. Academy of Management Journal 26 (June) 213-230. Hambrick, Donald C.2003. On the staying power of defenders, analyzers, and prospectors. working paper. Hennig, Christian. 2000. Identifiability of models for clusterwise linear regression. Journal of Classification 17 273-296. Hitt, Michael A. and R. Duane Ireland. 1986. Relationships among corporate level distinctive competencies, diversification strategy, corporate structure and performance. Journal of Management Studies 23(4) 401-416. Hosmer, D.W. 1973. A comparison of iterative maximum likelihood estimates of the parameters of a mixture of two normal distributions under three different types of samples. Biometrics 29 761-770. Hosmer, D.W. 1974. Maximum likelihood estimates of the parameters of a mixture of two regression lines. Communications in Statistics 3 995-1006. James, W., K. Hatton. 1995. Further evidence on the validity of the self-typing paragraph approach: Miles and Snow strategic archetypes in banking. Strategic Management Journal 16(2) 161-168. Jarvenpaa, Sirkka. 1989. The effect of task demands and graphical format on information processing strategies. Management Science 35(3) 285-303. Jaworski, Bernard J., Ajay K. Kohli. 1993. Market orientation: Antecedents and consequences. Journal of Marketing 57(3) 53-70. Jedidi, Kamel, Harsharanjeet Jagpal, Wayne DeSarbo. 1997a. Semi-parametric structural equation models for response-based segmentation and unobserved heterogeneity. Marketing Science 16 39-59.

31

Jedidi, Kamel, Harsharanjeet Jagpal, Wayne DeSarbo. 1997b. STEMM: A general finite mixture structural equation model. Journal of Classification 14 23-50. Koehler , A. B., E. S. Murphee. 1988. A comparison of the Akaike and Schwartz criteria for selecting model order. Applied Statistics 37 187-195. Lawson, Charles L.,Richard J. Hanson. 1972. Solving Least Squares Problems, Englewood Cliffs, NJ: Prentice Hall, Inc. Mahoney, John T., J. Rajendran Pandian. 1992. The resource-based view within the conversation of strategic management. Strategic Management Journal 13 363-380. McDaniel, Stephen W. ,James W. Kolari. 1987. Marketing strategy implications of the Miles and Snow strategic typology. Journal of Marketing 51(4) 19-30. McKee, Daryl L., P. Rajan Varadarajan, William M. Pride. 1989. Strategic adaptability and firm performance: A market-contingent perspective. Journal of Marketing 53(3) 2135. Meilijson, I. 1989. A fast improvement to the EM algorithm on its own terms. Journal of the Royal Statistical Society 51 127-138. Miles, R.,C. Snow. 1978. Organizational Strategy, Structure, and Process. New York: McGraw-Hill. Moenaert, Rudy K.,William E. Souder. 1996. Context and antecedents of information utility at the R&D/marketing interface. Management Science 42(11) 1592-1610. Narver, J.C., S.F. Slater. 1990. The effect of a market orientation on business profitability. Journal of Marketing 54 20-35. Porter, Michael. 1980. Competitive Strategy. New York: Free Press. Powell, M., J. D. 1977. Restart procedures for the conjugate gradient method. Mathematical Programming 12 241-254. Rajaratnam, Daniel, Lawrence B. Chonko. 1995. The effect of business strategy type on marketing organization design, product-market growth strategy, relative marketing effort, and organization performance. Journal of Marketing Theory and Practice 3(3) 60-75. Robinson, William T., Claes Fornell, M. W. Sullivan. 1992. Are market pioneers intrinsically stronger than later entrants? Strategic Management Journal 13 (November) 609-624.

32

Ruekert, Robert W., Orville C. Walker, Jr. 1987. Interactions between marketing and R&D departments in implementing different business strategies. Strategic Management Journal 8 233-248. Schwarz, G. 1978. Estimating the dimension of a model. Annals of Statistics 6 461-464. Segev, E. 1989. A systematic comparative analysis and synthesis of two business-level strategic typologies. Strategic Management Journal 10 487-505. Shortell, Stephen M., Edward J. Zajac. 1990. Perceptual and archival measures of Miles and Snow’s strategic types: A comprehensive assessment of reliability and validity. Academy of Management Journal 33(4) 817-832. Song, X. Michael, Mitzi M. Montoya-Weiss. 2001. The effect of perceived technological uncertainty on Japanese new product development. Academy of Management Journal 44 61-80. Swanson, E. Burton, 1994. Information systems innovations among organizations. Management Science 40(9) 1069-1092. Teicher, H. 1961. Identifiability of mixtures. Annals of Mathematical Statistics 32 244248. Teicher, H. 1963. Identifiability of finite mixtures. Annals of Mathematical Statistics 34 1265-1269. Titterington, D.M., A. F. M. Smith, U. E. Makov. 1985. Statistical Analysis of Finite Mixture Distributions. New York: Wiley. Walker, Orville C., Jr., Harper W. Boyd, Jr., John Mullins, Jean-Claude Larréché. 2003. Marketing Strategy: Planning and Implementation, fourth edition. Homewood, IL: Irwin/McGraw-Hill. Wedel, Michel and Wayne S. DeSarbo (1994), “A review of recent developments in latent class regression models.” In: Advanced methods of Marketing Research, (Richard P. Bagozzi, Ed.), New York: Blackwell, 1994, 352-388 Wedel, Michel and Wagner A. Kamakura (1999), Market Segmentation: Conceptual and Methodological Foundations. (Second Ed.). Boston: Kluwer. Wernerfelt, Berger. 1984. A resource-based view of the firm. Strategic Management Journal 5 171-180. Wu, C. F. J. 1983. On the convergence properties of the EM algorithm. Annals of Statistics 11 95-103.

33

Yakowitz, S. J. 1970. Unsupervised learning and the identification of finite mixtures. IEEE Transactions on Information Theory and Control IT-16 330-338. Yakowitz, S. J., J. D. Spragins. 1968. On the identifiability of finite mixtures. Annals of Mathematical Statistics 39 209-214. Yung, Y. F. 1997. Finite Mixtures in confirmatory factor analysis models. Psychometrika 62,, 297-330. Zajac, E., S. M. Shortell. 1989. Changing generic strategies: Likelihood, direction, and performance implications. Strategic Management Journal, 10 413-430. Zhu, H. T., S. Y. Lee. 2001. A Bayesian analysis of finite mixtures in the LISREL model. Psychometrika, 66 133-152.

34

TABLE 1 THE AGGREGATE COMPLETE SAMPLE SOLUTION

Capabilities

Profit

ROI

b

Intercept Marketing Technology Market Linking Information Technology Management Error Variance

0 0.162 0.436 0.156 0.285 0.021 0.658

Error Covariancec Error Correlation a All estimates in bold are significant (p