Vol. 37, No. 2, pp. 481-499
A Multivariate Model and Analysis of
Competitive Strategy in the U.S.
Hardwood Lumber Industry
ROBERT J. BUSH
STEVEN A. SINCLAIR
Business-level competitive strategy businessin the hardwood lumber industry was modeled
through the identification of strategic groups among large U.S. hardwood lumber pro-
ducers. Strategy was operationalized using a measure based on the variables developed
by Dess and Davis (1984). Factor and cluster analyses were used to define strategic
groups along the dimensions of cost leadership, focus,
A five strategic
group model was identified and examined for strategic orientation and intragroup homo-
geneity. Two groups had no distinctive strategic orientation that suggested a competitive
advantage; one group exhibited an overall cost leadership strategy; one exhibited a
differentiation strategy; and one group exhibited a dual overall cost leadership/
differentiation strategy. Strategic change in the industry was predicted to be primarily
toward increased differentiation. Three strategic groups indicated significant change in
this direction, and one group indicated an increase along both the focus
dimensions. FOR. SCI. 37(2):481-499.
ADDITIONAL KEY WORDS. Industry structure, strategic groups, strategy.
THE STRATEGIC GROUP CONSTRUCT is a relatively new and useful tool for
modeling and analyzing industries (Porter 1980, McGee and Thomas
1986). The underlying premise of the construct is that companies within an
industry are not necessarily homogeneous, but neither are all companies unique.
Instead, groups of similar companies can be defined such that the groups are, in
general, homogeneous within and heterogeneous between. Differences between
groups of firms are thought to be the result of deliberate strategic decisions and
thus reflect differences in strategic orientation (McGee and Thomas 1986). Stra-
tegic groups in this context can be defined as groups of companies that follow
similar competitive strategies (Harrigan 1985a).
The strategic group construct provides an important intermediate level for
industry analyses (Porter 1980). Studies of an industry as a whole may miss
important intraindustry strategic differences, and company-level analyses may not
generalize to the industry level (Harrigan 1983, Hatten and Hatten 1987).
The importance of the strategic group construct lies, primarily, in the relation-
hip of strategic groups to industry competition and performance. McGee and
Thomas (1986 p. 142) state that strategic groups, if they exist within an industry,
“. . . clearly have implications for the patterns of competition.” The complexity of
the strategic group structure within an industry has a significant influence on
economic performance (Newman 1978) and has been positively correlated with
industry competitiveness (Harrigan 1980, O’Laughlin and Ellefson 1981c, Hergert
1987). Strategic groups may also differ in their response to market opportunities
and threats (Thomas and Venkatraman 1988) and in their profit potential (Porter
1980). These relationships clearly suggest that the investigation of strategic
groups within an industry can provide important and useful information.
This paper reports the findings of a study that sought to improve understanding
of the U.S. hardwood lumber industry through the modeling of competitive strat-
egy among large hardwood lumber producers. Strategic groupings were used as
the basis of this model and as a framework for the prediction of strategic change
within the industry.
Hardwood lumber producers constitute an important segment of U.S. wood-
based industries. Luppold and Dempsey (1989) estimate that hardwood lumber
accounts for approximately one-third of the value of domestically produced lum-
ber—both hardwood and softwood. The industry’s importance is also suggested
by its employment of approximately 21,200 workers (USDC-BOC 1985) and its
position as supplier to high value-added industries such as household furniture and
Previous studies have investigated several aspects of the U.S. hardwood lum-
ber industry. Examples include: Luppold (1984), Abt (1987), Greber and White
(1982), and Buongiorno and Lu (1989). Luppold (1984) identified factors affecting
market growth and prices. Abt (1987) investigated factor demand using data for
Appalachian hardwood lumber companies. Greber and White (1982) and Buon-
giorno and Lu (1989) examined productivity in wood products industries, including
Many previous studies are limited, however, in that they implicitly treat the
industry as homogeneous. This assumption, while often necessary, may be inac-
curate. Company-specific resources and goals can result in differences in re-
sponses to exogenous factors and in company strategy. Strategic group analysis
addresses this latter difference and is the focus of this paper. Such analysis can aid
in understanding the industry and facilitate predictions of future industry changes.
In addition, empirical analysis of strategic groups aids in deterrnining the applica-
bility of theoretical strategic typologiesl to the industry.
THE STRATEGY CONSTRUCT
Strategies can be conceptually classified along two dimensions. The first of these
dimensions involves the corporate-level/business-level dichotomy. Business-level
refers to that level in an organization at which strategy for a single industry or
product market is determined (Hofer 1975). In contrast, corporate-level refers to
the top level of an organization, regardless of its size, and is concerned with the
configuration, management, organization, and financial transactions of business
units which typically operate in several industries (Patel and Younger 1978).
1Strategic topologies are classification systems that seek to simplify the concept of strategy to a
small set of generic strategies—strategies that apply generally regardless of industry, organization
type, or size (Herbert and Deresky 1987). Examples include the topologies proposed by Porter
(1980), Day (1984), and Chrisman et al. (1988).
Business-level strategy can be further classified along the second dimension—
strategic focus. Business-level strategies are comprised of investment, political,
and competitive substrategies (Hofer and Schendel 1978) each of which repre-
sents an area of strategic focus. Investment substrategies address the question of
optimum allocation of limited (primarily financial) resources, while political sub-
strategies are concerned with the firm’s interactions with external groups. Busi-
ness-level competitive substrategies (the focus of this research) address the
problem of achieving and maintaining a competitive advantage within an industry
or product market (Porter 1980, Day 1984, Patel and Younger 1978).
STRATEGY RESEARCH IN WOOD-BASED INDUSTRIES
Several studies have investigated business- or corporate-level strategy within
wood-based industries. Rich (1986) studied the corporate-level intended compet-
itive strategies of large wood-based corporations. The sample included 42 of the
largest U.S. corporations whose primary business was either fiber- or wood-
based products. Corporations were classified as using one or a combination of
Porter’s (1980) generic strategy types (overall cost leadership, differentiation, or
focus).2 Rather than infer overall corporate strategy from measurements of var-
ious strategic dimensions, Rich had respondents indicate directly which generic
strategy type their company employed.
In general, Rich (1986) found that the majority of firms reported utilizing an
overall cost leadership strategy. However, there was a trend toward the use of
differentiation and focus strategies when compared to the results of a similar study
covering the 1976-1979 period (Rich 1979). Firms utilizing differentiation and
focus strategies were also found to be more profitable during 1984 than firms
utilizing an overall cost leadership strategy.
In their three-article series, O’Laughlin and Ellefson (1981 a,b,c) examined the
structure of a multi-industry group of firms that included manufacturers of lumber
and lumber products (primarily softwood), paper and pulp products, and wood
household furniture. The sample consisted of the 40 largest (by 1978 sales rev-
enue) firms in this multi-industry group.
The third of O’Laughlin and Ellefson’s articles (1981c) examined strategic
groups within their sample. Firms were empirically classified into four strategic
groups based on 1978 sales rank and apparent diversification strategy. O’Laughlin
and Ellefson (1981c) concluded that competition in an industry increases with
increased strategic group complexity (i. e., the number of significant strategic
A more recent two-article series by Cleaves and O’Laughlin (1986a,b) exam-
ined realized business-level strategy within a sample consisting of 24 southern
pine plywood producers. Fourteen variables were measured for each of the 24
companies, and a hierarchical clustering algorithm was used to define five stra-
2Porter (1980) defines the strategies as follows: overall cost leadership requires that the firm seek
to become the industry’s low-cost producer without ignoring quality and service. A firm pursuing a
differentiation strategy seeks to produce a product or service that is perceived industry-wide as being
unique. Finally, a focus strategy requires that the firm concentrate on a particular market segment
and, in doing so, serve the segment more effectively or efficiently than can less specialized compet-
tegic groups. This methodology differed from that used by Rich (1986) in that it
did not force sample firms into previously defined strategy types.
Cleaves and O’Laughlin (1986a,b) point out that multidimensional clustering (as
used in their study) helps to explain competition among firms that cannot be
explained adequately using traditional economic models. In addition, the authors
suggest that the identification of strategic groups within an industry aids in pre-
dicting industry-wide response to government regulations, technological ad-
vances, changes in raw materials, and competitor moves.
The business-level competitive strategies utilized by U.S. pulp and paper prod-
ucts companies were studied by Bauerschmidt et al. (1986). In this study, 363
companies or business units rated the importance of 27 competitive methods.
Factor analysis of these ratings was used to empirically define five strategy types.
The fist two of these strategy types (differentiation and overall cost leader-
ship) are analogous to Porter’s (1980) generic strategies. The remaining three
strategy types are variations of Porter’s focus generic strategy. These included
product focus, geographic focus, and customer focus.
Bauerschmidt et al. (1986) concluded that the largest companies within the
sample competed exclusively on a cost basis, while smaller companies utilized one
of the three focus strategies. They point out that a differentiation-based strategy
can be both risky and difficult to implement in a commodity industry.
OPERATIONALIZING THE STRATEGY CONSTRUCT
Operationalization concerns the assignment of numbers to represent quantities of
attributes (Churchill 1979). In the case of business-level strategy, operationaliza-
tion requires that a complex phenomenon be simplified to a relatively small set of
measurable strategic dimensions.3 Researchers have found this process ex-
tremely difficult to accomplish in a manner that is consistent and widely applicable
(Hambrick 1980, Harrigan 1983). Numerous approaches have been investigated,
but none has been universally accepted. However, the choice of strategic dimen-
sions is extremely important since it is the single most influential factor in the
outcome of a study of strategy and the greatest source of variation between
studies (McGee and Thomas 1986).
Thomas and Venkatraman (1988) classify measurement schemes as narrow
(unidimensional) or broad (multidimensional). Narrow schemes use a single vari-
able such as company size, degree of vertical integration, or market share to
operationalize strategy. Broad schemes are based on observable characteristics of
the firm or scores on measures of various strategic dimensions.
The validity of narrow measurement schemes is limited since, at best, such
schemes can only be considered useful proxies that are correlated with strategy.
Thomas and Venkatraman (1988 p. 539) state the problem succinctly:
Our position is that the development of strategic groups using a narrow conceptu-
alization of strategy is unlikely to capture the complexity of the strategy construct,
thus limiting the usefulness of strategic groups for both descriptive and predictive
3As used in this paper, the term strategic dimension
refers to the major strategic directions in which
companies can move in order to gain a competitive advantage.
Because of this limitation, a multidimensional approach was used in this study.
Since strategy is a universal rather than industry-specific phenomenon, valid
measures should generalize across industry boundaries. Porter (1980) supported
the universality of strategy by proposing three generic strategy types that are
applicable to all industries. Dess and Davis (1982, 1984) built on Porter’s work by
developing and testing a measure of strategy that uses these three generic strat-
egy types as dimensions of overall strategic posture. This study adopted Dess and
Davis’ approach and utilized an adapted version of their measure.
The study focused on business-level intended competitive strategy within the
hardwood lumber industry. Since the strategy professed by company executives
may differ from the strategy that a company actually implements, intended strat-
egies may differ from realized strategies (Snow and Hambrick 1980). Realized
strategies may be the result of deliberate strategic decisions (intended strategies)
or they may reflect reactions to industry changes that have no underlying strategic
basis. However, focusing on intended strategy allows the use of strategic self-
typing by top management personnel. The perceptions and opinions of this group
largely determine the organizations strategy (Snow and Hambrick 1980). Focus-
ing on intended strategy also allows strategic change in the industry to be pre-
Quantitative strategic data were gathered via a 20-item measure adapted from
Dess and Davis (1982, 1984). The measure developed by Dess and Davis was
judged to have met the three concerns in strategic measurement presented by
Thomas and Venkatraman (1988): (1) It captured (with minor changes) the basis
of competition in the industry; (2) It had a strong relationship to existing strategic
topologies-specifically, Porter (1980); and (3) the works of Dess and Davis
(1982, 1984) provide evidence of the validity and reliability of the measure.
Minor changes were made in the measure to ensure applicability to the hard-
wood lumber industry. The content validity of the resulting measure was checked
through a review by knowledgeable Forest Service, university, and trade asso-
ciation personnel. Figure 1 lists the variables included in the measure.
The measure was incorporated into a questionnaire that also included questions
concerning the nature of the firm (sales, production levels, location, etc.). The 20
items included in the measure were rated for their importance to the company’s
competitive strategy using 7-point Likert-type scales that ranged from, 1 = Not
Important to 7 = Extremely Important. The questionnaire also asked recipients
to indicate how important they expected each item to be in their company’s future
(next 5 years) competitive strategy.
The sample used in this study consisted of the 100 largest (by production) U.S.
hardwood lumber producers. Sample companies were identified through a review
1.Developing new products2.Providing customer service3.Efficient operation of production facilities4.Product quality control5.Employing trained/experienced personnel6.Competitive pricing7.Developing brand identification8.Using new marketing techniques/methods9.Controlling channels of distribution10.Procurement of raw materials11.Serving special geographic markets12.Ability to manufacture specialty products13.Promotion and advertising14.Maintaining a company sales force15.Owning timberlands and/or logging operations16.Providing rapid delivery17.Market research18.Investment in new processing equipment19.Serving particular customer groups20.Reputation within the industry
FIGURE 1. Variables used to measure business-level strategy (adapted from Dess and Davis 1984).
of production figures provided by industry fact books (Miller Freeman 1987,
1988), trade association membership directories, The Weekly Hardwood Review
(Barrett 1987), and telephone conversations with company personnel. Where
companies participated in more than one industry, only the business unit involved
in hardwood lumber production was included in the study.
This nonprobabalistic approach to company selection resulted in a purposive
(judgment) sample and limits traditional probability-based extrapolations of the
study results to the entire industry. However, it was felt that given limited
research resources, strategic change within the industry could best be investi-
gated by examing larger, influential firms. Purposive sampling also allowed the
sample to be controlled for the potentially confounding effects of extreme varia-
tions in company scope and resources (Dess and Davis 1984). Nonprobability
samples are commonly used in marketing research (Green and Tull 1978). In
addition, Karmel and Jain (1987) have shown that a nonrandom, purposive sample
of large firms within an industry can outperform randomized sampling schemes for
estimating industry parameters.
Survey techniques were used to gather data from the sample firms. In multi-
industry companies, the questionnaire was directed to the head of the business
unit producing hardwood lumber. In single-industry companies, the questionnaire
was directed to the top executive. In some cases, it was not possible to contact
the top executive, and senior marketing/sales people were substituted.
The questionnaire was mailed during June 1989 to 80 sample companies. An
additional 19 questionnaires (one sample firm refused to be interviewed) were
administered between June and September 1989 as part of in-person interviews.
A total of 72 questionnaires (72%) were returned by the time analysis began.
IDENTIFYING STRATEGIC GROUPSFactor Analysis
Factor analysis refers to a group of multivariate methods for establishing dimen-
sions within a data set and for data reduction (Stewart 1981, Hair et al. 1987). In
this study, principal-component (factor) analysis was used to confirm the measure
used to operationalize strategy and to generate factor scores for use in cluster
Factor analysis of the competitive variable ratings requires that the data be
considered interval-scaled (Norusis 1988). Depending on the assumptions one
makes, rating-scale data can be considered to be ordinal, interval, or ratio-scaled
(Green and Tull 1978). While some authors have expressed concern with the use
of metric statistics with rating-scale data (see, for example, Martilla and Carve y
1975), such use is generally accepted in the marketing and strategic management
literature and was followed in this study.
Factor analysis was deemed an appropriate technique since examination of the
correlation matrix (Table 1) suggested relationships between variables, and a
Bartlett test of sphericity (Stewart 1981) rejected the hypothesis that the matrix
was an identity (P < 0.000). In addition, the Kaiser—Meyer—Olkin measure of
sampling adequacy (O. 71) was within the range considered acceptable by Stewart
(1981) and Norusis (1988).
The three-factor solution was chosen a priori
since the measure was designed
to evaluate Porter’s (1980) three generic strategies as dimensions of competitive
strategy. In addition, the three-factor solution was supported by a scree test and
an examination of factor eigenvalues (Stevens 1986). Table 2 provides the result-
ing factor loadings after Varirmax (orthogonal) rotation.
Recommendations vary as to the level at which a factor loading can be consid-
ered significant. Hair et al. (1987) report that, as a rule of thumb, factor loadings
with an absolute value greater than 0.30 can be considered significant. Stevens
(1986) suggests that only loadings with an absolute value greater than 0.40 have
practical significance. In keeping with Stevens’ more conservative recommenda-
tion, variable 15 (ownership of timberlands and/or logging operations), which was
designed to assess the importance of backward integration, was excluded from
further analyses due to its low loading on all three factors.
The remaining variables were assigned to the factor on which they had the
greatest loading and formed submeasures that represented the three strategic
dimensions. The reliability of the submeasures was evaluated by computing Cron-
back’s Alpha, a commonly accepted formula for assessing the reliability of a
multi-item measure (Peter 1979). Table 2 provides the Alpha values for each
submeasure. These values are considered acceptable by Churchill (1979) for
Analysis of the variables that were assigned to Factor 1 indicated that it clearly
Correlation matrix of variables used to operationalize strategy.
1Variable numbers reference Figure 1.
Factor and submeasure structure after varimax rotation.
. . . . . . . . . . . . . . . . . . . . . . Factor loadings1
. . . . . . . . . . . . . . . . . . . . . .
Using new marketing
Developing brand identification
Serving special geographic markets
Promotion and advertising
Serving particular customer groups
Ability to manufacture specialty
Developing new products
Maintaining a company sales force
Controlling channels of distribution
Owning timberlands and/or
Providing customer service
Product quality control
Efficient operation of production
Reputation within the industry
Providing rapid delivery
Procurement of raw materials
Investment in new processing
(For items forming submeasure)
1Bold type denotes the variables used to form the factor submeasure.
represented the differentiation dimension. Factor3 represented the cost leader-
ship dimension. Interpretation of Factor 2 was less clear since it incorporated
variables that were originally thought to assess either the differentiation or cost
leadership dimensions. This result is not surprising since a focus strategy, as
defined by Porter (1980), is a differentiation and/or overall cost leadership strat-
egy aimed at a specific market segment. Consequently, Factor 2 was interpreted
as representing the focus dimension.Cluster Analysis
Cluster analysis is a term applied to a group of empirical techniques for classifi-
cation of objects without prior assumptions about the population (Punj and Stewart
1983). While developed in the biological sciences, cluster analytic techniques are
commonly used in marketing research (Saunders 1980).
In this study, hierarchical agglomerative cluster algorithms were used to de-
termine strategic groupings among the sample companies. Factor submeasure
scores for each company were generated for input into the cluster algorithm using
the model:F = a x + a x + a x . . . a x
(1)i i1i1 i2i2 i3i3 ikik
Score on submeasure i (i
= 1 to 3)i a
= Rating of the importance on the first variable included in submeasure ii1x
= Rotated factor loading of variable a
on factor ii1i1 k
= Number of variables included in the submeasure
Prior to clustering, the data were examined for the presence of potential
outliers that could skew the cluster solution. Based on plots of the three sub-
measure scores and the Mahalanobis distance statistic (Norusis 1988) for each
company, two outliers were identified and removed from further analyses. An
additional two companies were removed due to missing data—resulting in a clus-
ter sample that included 68 companies.
The companies were first clustered using Ward’s method, which seeks to
minimize the sum of squared within-cluster distance (Hair et al. 1987). This
algorithm was chosen because it has been shown to outperform others in many
situations (Punj and Stewart 1983) and is the most conceptually appealing for the
identification of strategic groups.
Unlike theoretical statistics, cluster analysis does not provide precise rules for
choosing a solution (Dess and Davis 1984, Harrigan 1985a). Instead, the choice of
an appropriate solution must be based on less rigid guidelines and the interpret-
ability of the results. A five-cluster solution was chosen based on analysis of a plot
of the number of clusters versus the standardized distance coefficient and because
this number of clusters was the smallest that adequately differentiated the com-
The reliability of the five cluster solution was tested using the three part
approach suggested by Choffray and Lilien (1980) and used by Doyle and Saun-
ders (1985). This approach consists of:
1. Testing for outliers in the data.
2. Testing the randomness of the data structure.
3. Testing the uniqueness of the solution.
The first of these tests has been previously described. To facilitate testing the
randomness of the data structure, 15 sets of random data with distribution char-
acteristics (mean and standard deviation) similar to the actual data were gener-
ated. Each of these data sets was clustered using Ward’s method. The mean
standardized distance coefficients at critical cluster levels were then compared to
the coefficients from the actual data (Table 3). If the distance coefficient did not
significantly differ from the random data, the cluster solution would be trivial. As
illustrated in Table 3, significant differences were noted—suggesting that an
underlying structure existed in the data.
The final test required that the cluster solution based on Ward’s method of
cluster formation be compared to the solutions based on alternative methods. This
- THEORY AND PREVIOUS RESEARCH
- The Strategy Construct
- Strategy Research in Wood-Based Industries
- Operationalizing the Strategy Construct
- Research Instrument
- Data Collection
- Identifying Strategic Groups
- Factor Analysis
- Cluster Analysis
- Strategic Groups
- Competitive Advantage
- Group Profiles
- Group Homogeneity
- Strategic Change
- LITERATURE CITED
- AUTHORS AND ACKNOWLEDGMENTS