FIRM-SPECIFIC COGNITIVE FRAMES AS CO-DETERMINANTS OF
PERSISTENT PERFORMANCE DIFFERENTIALS
By
THORBJØRN KNUDSEN
University of Southern Denmark, Main campus: Odense University
Department of Marketing
55 Campusvej,
DK-5230 Odense M, Denmark
Tel.: +45 6650 3148
Fax: +45 6615 5129
E-mail: [email protected]
Draft of 29 August, 2001
I am grateful for comments from and discussions with Suzanne Beckmann, Richard M. Burton, Bo Eriksen, Hans
Rask Jensen, Jørgen Lauridsen, Michael Lubatkin, Tage Koed Madsen and James G. March. All remaining errors
were produced without any help.
1
FIRM-SPECIFIC COGNITIVE FRAMES AS CO-DETERMINANTS OF
PERSISTENT PERFORMANCE DIFFERENTIALS
ABSTRACT
The present study extends the Resource Based View by appealing to firm-specific cognitive
frames as co-determinants of persistent performance differentials. It is proposed that cognitive
frames shared among members of a specific business organization influence financial
performance by a causal chain running through information processing and uncertainty. The
empirical test supports the proposed relation between cognitive frames and performance. The
evidence is based on survey data obtained in 1995, a follow-up survey conducted 1999 and
archival data. Furthermore, the empirical analysis introduces a novel mediator test to estimate
path models when the usual structural equation approach is infeasible.
Key words: Cognitive frames, Information processing, Uncertainty, Resource Based View,
Mediator test.
Running Header: Cognitive frames and performance differentials
2
FIRM SPECIFIC COGNITIVE FRAMES AS CO-DETERMINANTS OF
PERSISTENT PERFORMANCE DIFFERENTIALS
The literature on organizational cognition and learning has suggested that an important
feature of training new organization members is the socialization process, which instills a set
of beliefs and cognitive frames that are shared among all employees of a particular
organization (Brown et al., 1989; Eden & Spender, 1998; Nonaka & Takeuchi, 1995). In turn,
shared cognitive frames are drawn upon to produce meaning by enriching incoming sense data
and to validate and legitimize structural relations among organization members (Ranson et al.,
1980). By cognitive frames, I refer to what Bartunek (1984), Giddens (1979, 1984) and Schutz
(1967) defined as interpretive shemes. Thus, in the present paper, cognitive frames are
categories “…which map our experience of the world identifying both its relevant aspects and
how we are to understand them…” (Bartunek, 1984: 355). Once they are established,
individual cognitive frames may, according to the literature on cognition and decision-making,
act as a “reference point” that provides a mental frame for subsequent selection and evaluation
of information (Minsky, 1975; Tversky & Kahneman, 1981).
In the organizational literature, the shared aspect of cognitive frames has been
emphasized (Eden & Spender, 1998; Pfeffer, 1997). Through socialization, (individual)
cognitive frames will be shared in the sense that they refer to experience acquired in a common
organizational world. Tsoukas (1996) used the term “dispositions” for this aspect of
individually held knowledge formed in the course of past socialization. Other terms with a
similar meaning include cognitive maps (Weick & Bougnon, 1986), interpretive schemes
(Schutz, 1967; Giddens, 1979; Bartunek, 1984), contextual schemes (Shutz, 1967),
institutional frames (Elliott et al., 1998), paradigms (Kuhn, 1970; Pfeffer, 1997), world-views
and ideologies (Beyer, 1981; Starbuck, 1982) as well as organizational culture (Jelinek et al.,
1983). Despite important variations in emphasis, these concepts have the following aspects of
meaning in common: (1) they refer to generic cognitive frames that underlie all aspects of
human knowledge and skill, (2) their processing lies beyond the direct reach of awareness but
their products (words, images, emotions and actions) are directly available, (3) they are
3
relatively stable, partly because their processing lies outside the reach of consciousness, (4)
they play a crucial role in engendering commitment among organization members and shaping
the problems organization members will, and will not, address, (5) they are individually held,
hence any given cognitive frame may have multiple meanings, yet (6) individual cognitive
frames are socially shaped. The present paper will understand the term “cognitive frame” to
hold these aspects of meaning.
Pfeffer (1997) has argued that there is an open question regarding the possible sources
of organization-wide social influences upon individual cognition. The process of socializing
new members into a particular organization provides one answer to this question. Through
socialization, new organization members gradually acquire a common set of cognitive frames
and beliefs (Nonaka & Takeuchi, 1995; van Maanen, 1973). Since the socialization process
takes time and relies on prolonged face-to-face interaction with other organization members,
including organizational heroes and other role models, the new organization member acquires
beliefs and cognitive frames, which are shared with other organizational members (Brown et
al., 1989; Nonaka & Takeuchi, 1995; van Maanen, 1973). That is, cognition is situated within a
specific organizational or social context (Brown et al., 1989). And the more thorough the
socialization process, the more deeply ingrained the firm- or organization specific cognitive
frames and beliefs. An extreme example is the former U.S. Navy SEAL members’ recollection
of the organization and team-specific training they received and its enduring effect (Dockery &
Fawcett, 1998). Another striking example of the ingrained nature of beliefs and cognitive
frames is provided by the failure to change an organizational culture despite its recognised
negative impact on the members’ stress levels (Perlow, 1999).
In the strategy literature, proponents of the Resource Based View (RBV henceforth)
argue that firm-specific immobile factors, combined with uncertainty, are crucial determinants
of persistent performance differences between firms competing in the same industry (for a
good summary, see Peteraf, 1993). Although the argument for persistent performance
differentials favoured by the RBV relies on persistent heterogeneity of firm-specific (human)
capital, the link to the recent literature on organizational learning and cognition, which
4
provides support for this contention, remains underdeveloped. Previous studies have linked
interpretation to strategic adaptation (see e.g. Barr, 1998) but no explicit links between
cognitive frames and performance have been suggested. Since cognitive frames, aquired
through a prolonged socialization process, can be deeply ingrained and thus have an enduring
and stable quality characteristic of a particular organization, they are obvious candidates as codeterminants of firm-specific competitive advantage. Especially, since the expression of
cognitive frames depend on the enduring presence of other organization members sharing
similar frames, a fact which, makes the transfer of cognitive competence from one organization
to the next problematic (Nonaka & Takeuchi, 1995). In consequence, the present paper
provides a possible link between the literature on organizational cognition and the RBV,
suggesting that firm-specific cognitive frames are co-determinants of persistent performance
differentials. The proposed causal chain between cognitive frames and performance can be
summarised in five steps: (1) cognitive frames, by definition, exclude some information from
consideration by the organization, (2) this excluded information will increase uncertainty, (3)
the increase in uncertainty, on average, decreases financial performance, (4) this decrease in
financial performance may be off-set by an increase in information processing, and (5) because
cognitive frames are thought to be enduring, this impact on a firm’s financial performance will
persist over time. This argument is justified in the following section, including the crucial
assumption that cognitive frames are enduring.
The present paper consists of five sections. First, I present a general argument that links
firm-specific cognitive frames and financial performance through information processing and
uncertainty. Second, the choice of the particular cognitive frames used for the empirical test is
motivated (a strategic frame and a myopic frame related to environmental issues) and the
general theoretical argument is applied to formulate the hypotheses to be tested in the present
paper. Third, the novel estimation procedure developed to handle the estimation of the
theoretical model is presented (a four-step procedure for handling difficult path-models) along
with the data used for estimation. Fourth, I present the results of the empirical test. The final
and fifth section provides discussion and conclusion.
5
Linking Firm-Specific Cognitive Frames and Financial Performance
An obvious link between the literature on organizational cognition and the RBV is
provided by considering the role of cognitive frames in facilitating decision-making in an
uncertain world. Both theory and empirical evidence suggest that decision makers faced with
an uncertain world need a set of guiding principles in order to structure and coordinate the
firm’s actions (Kellaris et al., 1996, Levinthal & March, 1993; Rabin, 1998). The core of this
argument is well established and founded on the idea that cognitive frames provide a
simplifying heuristic used to manage the unlimited detail of everyday life (see e.g. Eden &
Spender, 1998; Rabin, 1998). In consequence, accomplishing simplification entails the
ignorance of some detail and thus loss of information. What specific information gets lost
depends on the particular frame used. Therefore, the role of a particular cognitive frame may
be seen as a screening device, which enables decision-making by identifying the important and
salient aspects of the firm’s environment (Bartunek, 1984). As a result, the unimportant detail
is ignored and limited attention is put to better use, an argument, which can be traced to
Simon’s (1955) classic paper on bounded rationality.
Considering Simon’s (1955) argument, as well as traditional views in (economic)
psychology, which implies that cognition is an activity that takes place on the individual level,
it is clearly problematic to reify the organization as a cognizing entity (Eden & Spender, 1998).
It is plausible, however, to argue that cognitive frames exist as organization-level phenomena
in the sense that they come about through socialization, persist through social interaction and
are shared by organization members (Eden & Spender, 1998; Nonaka & Takeuchi, 1995).
Not only are cognitive frames shared among organization members and thus an
organization-wide phenomenon, also, information processing can be seen as an organizationlevel phenomenon. Thus, according to Egelhoff (1982: p.435-436), Galbraith (1973) and
Tushman & Nadler (1978), information processing in organizations is generally defined as,
“…including the gathering of data, the transformation of data into information, and the
communication and storage of information in the organization.” Although it is individuals,
which process information, they do so within a specific set of enduring organizational
6
interaction patterns involving coordination through recurrent use of oral communications,
policies, procedures, work plans, personal contact and meetings (Tushman, 1978; Van de Ven
et al., 1976; Egelhoff, 1982).
Focussing on the organization-level, the role of cognitive frames, which are shared
among organization members, can be seen as a device that directs attention towards those
features of the environment, which are important for the firm’s performance and viability. But
the possibility also exists that both individual and shared cognitive frames may be degenerate
(Reason, 1990). A particular cognitive frame may direct attention to the wrong information or
inspire consideration of too little or too much information etc. To the extent that cognitive
frames can be viewed as relatively stable firm-specific features of organization, they may,
therefore, be important co-determinants of firm-specific differences in information processing
and, ultimately, performance. The following paragraphs provide further detail regarding the
relative stability of cognitive frames.
The relative stability or rigidity of cognitive frames has been a subject of some debate.
Thus, Rosman et al. (1994) argue that agents cannot be portrayed as either rigid or flexible. It
all depends on the particular talent of the decision-maker, the task at hand, the situation, and
the context of the decision. Cognitive frames do change, but usually the change is a gradual
and slow incremental modification in present ways of interpretation (Bartunek, 1984).
Fundamental shifts in cognitive frames are much more rare than incremental change but when
they do happen, dramatical and wide-ranging effects on the organization’s structural form can
be observed (Bartunek, 1984; Tushman & Romanelli, 1985).
Furthermore, broader and more abstract cognitive frames are more resistent to revision
than narrow and specific frames (Kuhn, 1970; Minsky, 1975), a view, which emphasises the
embedded nature of cognitive frames (Reason, 1990). Abstract cognitive frames will contain
specific frames as subparts (Reason, 1990). Thus, an office-building frame will have office
frames as subparts and office frames will include desk, filing cabinet and computer frames as
sub-components (Reason, 1990). The point is that narrow and specific frames are much more
likely to change than the abstract and broad frames within which they are nested (Rumelhart,
7
1975; Reason, 1990). A further reason for the stability of cognitive frames noted in the above
was that their processing lies beyond the direct reach of awareness even if their products
(words, images, emotions and actions) are directly available (see Reason, 1990). Thus, an
underlying cognitive frame that cannot be directly reached by consciousness should be much
more difficult to change than a grocery list.
Having established that broad cognitive frames can be seen as relatively stable features
of particular organizations, the causal chain between cognitive frames and performance can
now be spelled out in more detail. The argument involves five steps.
First, cognitive frames enable a clearer conceptualization of one part of the world by
increasing information processing within the bounds defined by the specific cognitive frame.
But this benefit always comes at the cost of excluding other aspects of the world beyond the
bounds of a particular cognitive frame. Therefore, the use of a cognitive frame, per definition
excludes potentially valuable information.
Second, this excluded information will increase uncertainty. Thus, applying a cognitive
frame will increase actual uncertainty in the sense that some part of the distribution of
outcomes is simply not evaluated (Arrow, 1974; Levinthal & March, 1993). This effect should
be present for all types of cognitive frames since their function is to direct attention towards a
limited part of the world.
Third, the increase in uncertainty, on average, decreases financial performance. The
assertion that uncertainty (understood as risk) decreases the expected value of performance by
increasing the variance of outcomes is based on a standard argument in the literature on
decision-making (see Luce & Raiffa, 1957). An alternative version of the argument says that,
as the range of possible outcomes increases, it is increasingly difficult (and therefore costly) to
write contracts and to insure against adverse states of nature. Understanding uncertainty as a
situation where point probabilities cannot be estimated without further information, or not at
all (often referred to as Knightian uncertain), will imply a further decrease in average
performance compared to the situation of risk.
Fourth, the decrease in financial performance caused by uncertainty may be off-set by
8
an increase in information processing. Whereas average performance decreases in uncertainty,
information processing can be assumed to increase the expected value of performance by
providing better estimates of outcomes. This argument is a standard one in economics and
relies on the assumption that uncertainty in principle can be remedied by supplying more
information (Arrow, 1974; Luce & Raiffa, 1957). Obviously, the argument does not hold if
“bad” information is processed or if causal ambiguity prevents uncertainty to be relieved by
more information. In this case, increasing information processing will not help. Thus, it is an
open question whether uncertainty may be off-set by information processing.
Fifth, because cognitive frames are thought to be enduring, this impact on a firm’s
financial performance will persist over time. This argument should hold in the case of broad
cognitive frames, which can be seen as relatively stable features of particular organizations.
Against this argument may be raised the objection, why cognitive frames are not simply
changed when they are proven degenerate? Why did Facit, for example, persist in their
application of a type-writer frame long after it had become obsolete? As mentioned in the
above, there are at least two reasons why degenerate cognitive frames may endure: (1)
cognitive frames may simply be so deeply ingrained that they resist change despite the fact that
they have been recognised as degenerate in some sense (see Perlow, 1999 for an example), (2)
if cognitive frames influence performance indirectly through information processing and
uncertainty, it may be difficult to recognise the causal effect of a particular cognitive frame.
In sum, the above five-step argument establishes a causal chain in which cognitive
frames are sources of more effective use of limited attention by excluding some information. In
turn, the excluded information will increase uncertainty, and, uncertainty reduces performance.
Cognitive frames also influence information processing, which may off-set the negative
uncertainty effect. Now, since broad cognitive frames may be conceived as rather stable firmspecific frames of reference, it is established that at least some cognitive frames may have an
important role as co-determinants of persistent performance differentials.
Thus, the present paper adds to previous RBV arguments by identifying an explicit
chain of causality between cognitive frames and performance, and, on this basis, suggesting
9
how cognitive frames may be sources of persistent performance differentials among firms
which otherwise share similar industrial conditions. Note that the usual RBV argument hinges
upon resource immobility, demand- or supply-side uncertainty and resource heterogeneity (see
e.g. Peteraf, 1993). The present argument extends this line of reasoning by suggesting how
specific causes for immobility and uncertainty (i.e. limited attention entailed by the use of
cognitive frames) by way of an explicit and testable causal chain may translate into
performance differentials. Although the theoretical argument is general, the empirical test is
limited to two cognitive frames (a strategic frame and a myopic frame) and is scoped within
the context of issues related to the natural environment. This choice is motivated in the ensuing
section.
Linking a Strategic and a Myopic Frame Related to Environmental Issues with Financial
Performance
In this section, the choice of the particular cognitive frames used for the empirical test
is motivated and the general theoretical argument is applied to formulate the specific
hypotheses to be tested in the present paper. I proceed as follows.
First, all cognitive frames involve simplification, loss of information, and so forth. But
only some cognitive frames may influence financial performance. Therefore, I first identify a
specific set of cognitive frames, which according to the literature are likely to have this
property (a strategic and a myopic frame). Second, since the empirical test involves the need to
operationalize uncertainty and information processing, it is necessary to select a particular
empirical context within which these constructs are salient and may be defined in a meaningful
way (firms which are significantly influenced by issues related to the natural environment).
Third, having selected a specific set of cognitive frames and narrowed the scope of the
empirical test to a case where uncertainty is a persistent feature of the context for the firm’s
decision-making, I proceed to formulate the hypotheses to be tested in the present study.
Selecting cognitive frames for the empirical test. In this section, I motivate the choice
of the particular cognitive frames to be used for the empirical test and establish their link to
uncertainty and information processing. I start by characterizing the two broad cognitive
10
frames, a strategic frame and a myopic frame, chosen for the empirical test due to their
ubiquity in theory and practice (see e.g. Levinthal & March, 1993; Porter, 1980; Rumelt et al.,
1995). The relation between these two cognitive frames and the scope of information
processing suggests they capture fundamental ideas of what business is about. Thus, the scope
of information processing is broad when a strategic frame directs decision-making and it is
narrow when guided by a myopic frame (Levinthal & March, 1993). Given the competence in
evaluating alternatives, differences in framing will influence the scope of information
processing. A myopic frame will tend to ignore effects distant from immediate antecedents of
efficiency (Levinthal & March, 1993). Therefore, a myopic frame economizes on sample size
of alternatives that are evaluated but will increase the number of potential surprises due to a
narrow sample of possible futures (Ibid.). By contrast, a strategic frame will, at the cost of
large samples of alternatives, tend to include distant effects and thus reduce potential surprises
(Ibid.). Moreover, an emphasis on immediate costs over uncertain long-term benefits is often
associated with a myopic frame (Porter, 1980). Observations within the context of utilization
of information technology lend support to this conjecture. Thus Fletcher & Wright (1995)
found that a general reliance on traditional cost-benefit appraisal methods for integration of
new information technology indicates a short-term, rather than a long-term strategic, focus for
information systems use. An even more dramatic example is the role of myopia as the root
cause of Food Lion’s failure (Dess & Picken, 1999). According to Dess & Picken (1999: 100),
organizational myopia “carried the strategy of cost cutting to a dysfunctional extreme.” As
indicated by this example, a myopic frame is closely associated with a narrow cost-focus and is
possibly non-adaptive.
Selecting the context for the empirical test. Even if the theoretical argument is general,
the empirical test is limited to the two cognitive frames “strategic frame” and “myopic frame”
and bounded within the context of environmental issues. That is, the hypotheses are bounded
within a sample of firms that are influenced by environmental issues. Therefore, the
operationalization of cognitive frames, information processing and uncertainty all concern the
firm’s consideration of such issues. Since our hypotheses are limited by these considerations, it
11
is necessary to motivate the scope of the empirical test before turning to the formulation of
hypotheses.
In order to test the hypotheses drawn on the above argument, I was looking for a case
where some uncertainty should be a persistent feature of the context for the firm’s decisionmaking. Although many settings were possible, I focussed upon firms which produced high
amounts of hazardous waste, and, therefore, were prone to influence by stakeholders
promoting various different issues related to the natural environment. In light of the last twenty
years public focus on firms’ behavior towards the natural environment, it is increasingly
important for managers to consider non-market stakeholders such as government agencies,
pressure groups and media. According to the literature on this subject (see e.g. the material
contained in Harrison & Freeman, 1999 and Fischer & Schot, 1993), issues related to the
natural environment thus introduce the presence of uncertainty necessary to test the hypotheses
summarized in the model shown in Figure 1, below.
Linking cognitive frames and performance within the bounds of the empirical test.
We now turn to formulating hypotheses that stipulate a possible relation between strategic
frames, myopic frames, information processing, uncertainty and financial performance.
Since the effects of environmental issues are usually distant from immediate efficiency
concerns, but relevant for strategic reasons (Klassen & Whybark 1999, Fischer and Schot,
1993; Hoffman, 1999; Shrivastava, 1992; Wolff, 1995), a strategic frame and a myopic frame
should have opposite effects on information processing. Since most non-market stakeholders
associated with environmental issues are irrelevant in terms of short-term cost and efficiency
considerations (see e.g. Fischer and Schot, 1993; Wolff, 1995), a myopic frame should
decrease information processing related to environmental issues. By contrast, a strategic frame
should increase information processing related to environmental issues since non-market
stakeholders often raise new issues, some of which may potentially influence the firm’s
performance in the long term.
H1.1
A strategic frame increases information processing related to issues, which
have an uncertain influence on the organization in the long term.
12
H1.2
A myopic frame decreases information processing related to issues, which have
an uncertain influence on the organization in the long term.
These hypotheses of are proposed on basis of the argument that cognitive frames are
persistent organization-level features and that information processing is an organization-level
phenomenon. The empirical test of hypotheses H1.1 and H1.2 is based on the assumption that
environmental issues primarily will have an uncertain long-term effect on business
organizations, an assertion which is supported by the literature on coporate greening (Klassen
& Whybark 1999; Shrivastava, 1992; Wolff, 1995). It should further be noted that H1.2
assumes that the minimum level of information processing has not yet been reached by the
organization. Since we study producers of large amounts of hazardous waste, it is possible that
a relatively high level of background information processing is required due to the necessary
interaction with authorities, waste treatment processors, workers and so on. If the required
background information processing is relatively high, a myopic frame may not reduce
information processing further. In case H1.2 should be refuted, a test of this assumption is only
possible on a sample, which includes firms that are not producers of hazardouos waste or
otherwise required to uphold a high level of background information processing. I test this
crucial assumption on data from a follow-up study conducted in 1999, as described below.
Furthermore, the above argument implies that cognitive frames guide the organization’s
information processing and, therefore, determine the sample size of the alternative outcomes
evaluated by the organization. Organizations with a strategic frame will consider the long term
whereas an organization with a myopic frame will consider the short term. Thus, a strategic
frame is associated with large samples of possible outcomes whereas a myopic frame implies
that more narrow samples are considered. Therefore, the number of situations to which point
probabilities are assigned to outcomes of actions will tend to be larger in organizations where a
strategic frame characterizes decision-making. In other words, the number of potential
surprises will decrease. When a myopic frame applies, the logic is reversed. On average, a
strategic frame should reduce uncertainty whereas a myopic frame should increase uncertainty.
In particular, uncertainty related to environmental issues should decrease in a strategic frame
13
and increase in a myopic frame (both related to environmental issues).
H1.3
A strategic frame reduces uncertainty.
H1.4
A myopic frame increases uncertainty.
Note that the empirical test of H1.3 and H1.4 are proposed under the assumption that
environmental issues are associated with the uncertainty that comes from complex stakeholder
interactions, an assumption, which is suggested in the literature (Harrison & Freeman, 1999;
Hoffman, 1999; Fischer & Schot, 1993). Moreover, it is assumed that the reason for reduced
uncertainty in the case of a strategic frame is that, a priori, fewer outcomes will be unknown
relative to organizations that use a myopic frame. This effect is caused by previous information
processing but does not involve information processing in the present period. Therefore,
cognitive frames are proposed to have a direct effect on uncertainty.
Considering information processing related to environmental issues, it is suggested that
a strategic frame will have two distinct effects. When a strategic frame is used, the sample size
of considered alternatives should be larger and also the evaluation of alternatives should
improve due to increased information processing intensity. Due to the combination of these
effects, a strategic frame is likely to have an indirect uncertainty reducing effect, which is
caused by the mediation of information processing. Hence, it is proposed that the negative
effect of a strategic frame on uncertainty will be mediated by information processing.
Similarly, the positive effect of a myopic frame on uncertainty will be mediated by information
processing.
H1.5
The negative relation between a strategic frame and uncertainty is mediated by
information processing.
H1.6
The positive relation between a myopic frame and uncertainty is mediated by
information processing.
Note that if a myopic frame is associated with an absence of information processing
directed towards environmental issues, there should be no mediation.
We next consider the relation between decision-making and outcomes. First, there is a
straightforward relation between uncertainty and performance. Increased uncertainty will on
14
average tend to reduce performance because uncertainty entails increased variance of outcomes
as well as costly adjustments following surprises (Luce & Raiffa, 1957). Second, information
processing improves alternative evaluation because it increases the correspondence between
the subjective and objective aspects of decision-making, i.e., between cognitive frames and the
regularities in the firm’s environment. Specifically, information processing increases the
correspondence between the actual and perceived importance of non-market factors such as
stakeholder claims. Apart from a hypothetical certain world, it is thus proposed that
incremental increases in information processing will also increase performance.
H2.1
Increased uncertainty reduces financial performance.
H2.2
Increased information processing increases financial performance.
Note that H2.2 implies a linear relation between information processing and
performance. It is proposed under the assumption that the hypothetical limit where the
marginal cost of information processing is higher than its marginal gain will not be reached in
practice. If this were the case, the relation would be curvi-linear. This question is clearly an
empirical one. As reported in the following, the data supported the assumption of a linear
positive relation between information processing and performance.
As mentioned above, information processing is understood as an active search process
where representations may be formed, and not merely updated. Hence, increased information
processing intensity entails cognitive frames that better correspond to actual events. Cognitive
frames that better correspond to actual events will be more accurate and perhaps cover a larger
portion of reality. Uncertainty has a double role here. As stated in H2.1, the direct effect of
uncertainty will on average decrease performance. Moreover, since uncertainty implies an
absence of conceptual frames for information processing, there is an indirect effect of
uncertainty, which also will decrease performance. This suggests that information processing
will have a negative mediating effect on the relation between uncertainty and performance.
H2.3
Information processing has a negative mediating effect on the relation between
uncertainty and financial performance.
Note that the argument concerning the relation between uncertainty, information
15
processing and performance suggests that the success in uncovering and adopting useful
alternatives critically depends on the cognitive frame that directs information processing.
Hypothesis H2.3 is proposed under the assumption that uncertainty reduces information
processing because the cognitive frames to guide information processing are absent. A less
severe situation would be the case where the necessary cognitive frames were in place. In this
case, H2.3 should be reversed. Whether the sign of H2.3 should be negative or positive is an
empirical question and will, therefore, be subject to the empirical test.
I have proposed that cognitive frames might influence information processing and
uncertainty but I do not find it plausible to suggest a general causal link between framing
effects and financial performance. Depending on the specificity of a particular cognitive frame
the link may be direct or indirect. The more specific a cognitive frame, the more likely will a
direct performance effect be and vice versa. As previously mentioned, the possibility of the
persistence also of degenerate cognitive frames should be greater if they are indirectly linked
to performance. If there was a direct link, the degenerate frame could much more easily be
identified and changed. Therefore, the proposed argument implies an indirect relation between
very broad firm-specific cognitive frames and performance mediated by information
processing and uncertainty. This leads to the following hypotheses of complete mediation
between broad cognitive frames and performance.
H3.1
The relation between a strategic frame and financial performance is completely
mediated by information processing and uncertainty.
H3.2
The relation between a myopic frame and financial performance is completely
mediated by uncertainty.
Note that the complete mediation in H3.1 and H3.2 are assumed on the basis that the
strategic and myopic frames are very broad. The more specific the cognitive frame, the more
implausible would the argument of complete mediation become.
The hypothesized relations are summarized in the path-diagram shown in Figure 1. As
can be seen, the three sets of hypotheses imply that a myopic frame will tend to decrease
financial performance due to an increase in uncertainty and an absence of information
16
processing. By contrast, a strategic frame entails a positive effect on performance. In sum, the
hypotheses establish a causal chain where cognitive frames are co-determinants of financial
performance, i.e., a number of other determinants not tested in the present study, are,for
example, firm-specific competences, network position and industrial conditions.
----------------------------------Figure 1 about here
----------------------------------An important concern is that the hypothesized effects may simply be too weak to stand
out from the clutter of all the other possible influences on performance. As previously
mentioned, the present study is scoped within the context of issues related to the natural
environment. Since firms that are affected by such issues are likely to experience a relatively
high level of uncertainty (e.g. caused by unexpected and conflicting stakeholder-demands), this
context is one in which the effects summarized in Figure 1 should be sufficiently strong to be
measurable. A possible drawback of this choice is that the influence of environmental
stakeholders is a fairly recent phenomenon, which implies that the tested cognitive frames
themselves may be transient. To examine whether this was the case, a follow-up study was
conducted in 1999, four years after the first survey was completed. The results from the
follow-up study are briefly summarized in the following (see appendix B).
There is one further snag associated with the model shown in Figure 1. Since it is
impossible to identify the model without imposing further constraints, it cannot be estimated
by a simultaneous equation approach to path-analysis. The solution to this problem is
presented in the following section.
METHODS
Estimation procedure
The hypothesized causal relations were summarized in the path diagram shown in
Figure 1. As previously noted, a structural equation approach to path-analysis is infeasible for
estimation of this model since it cannot be identified. Moreover, because the independent
17
variables are negatively correlated (see Table 2, below), the assumptions for such an approach
are violated. In consequence, we devised a novel estimation procedure in order to handle this
problem. Figure 2 shows the general procedure. Note that the diamond, in compressed form,
captures the relations shown in Figure 1.
----------------------------------Figure 2 about here
----------------------------------As shown, estimation of the model can be decomposed into a four-step estimation
procedure. The first and the second step in the procedure tests whether the two partial mediator
effects are present. The third step is necessary to test whether mediation between the
independent(s) and the dependent is complete or partial. Finally, the fourth step is needed to
establish whether the effect between the independent(s) and the dependent is in fact mediated
by the two intermediate variables. An obvious alternative explanation is that significant
mediator effects established in step one and two just represent variance which is “shuffled”
around. To reject this alternative, a significant effect in step four has to be shown. Furthermore,
the sign of the independent variable should approximately reflect the net effect on the
dependent variable according to the estimates obtained in step one and two.
On basis of the above four steps, models of the type shown in Figure 2 can be
estimated. Since yet larger models can be broken down into models of the type shown in
Figure 2, the method is general and lends itself readily to empirical tests in a number of
situations where structural modeling cannot be pursued. Obviously, there is a trade-off
between the requirements of theory and parsimony in the choice of estimation method. Even if
a structural equation approach should be used when applicable, there might be cases, as in the
present study, where theoretical considerations should override the constraints of a particular
estimation method. Having provided an overview of the estimation procedure, a brief account
of the estimation method is in order.
Estimation: step one and two. To establish mediation, estimation in step one is carried
out by three linear regression-equations (OLS-estimation) following Baron and Kenny’s
18
(1986) suggestions for differentiation of moderator and mediator effects of third variables.
Estimation in step two follows an identical procedure.
It is important to notice that mediation does not concern interaction (moderation). The
term “moderator effect” is equivalent to the commonly used term “interaction effect.”
Mediation is different, however. It is an intermediate mechanism that accounts for the relation
between the independent and the dependent variable.
To test for mediation we estimate a series of regression analyses where the path from
the independent to the dependent variable is controlled. When the effect of the independent
variable and the mediator variable are both significant, mediation holds if, further, the
independent variable has less effect when the mediator is controlled. This involves estimation
of three regression equations: (1) regressing the mediator on the independent variable, (2)
regressing the dependent variable on the independent variable, and (3) regressing the
dependent variable in both the independent variable and on the mediator (Baron and Kenny,
1986). Mediation is established when: (1) the effect of the independent variable is significant
in the two first equations, (2) the effect of the mediator is significant in the third equation, and
(3) the effect of the independent variable is less in the third equation. There is perfect
mediation when the independent variable has no effect when the mediator is controlled.
Estimation: step three. Estimated by linear regression (OLS-estimation).
Estimation: step four. Step four includes the model’s mediator variables (Z-variables
in Figure1) as instrumental variables (IV) in a two-stage least-squares estimation (the method
and the associated test-statistics are described in appendix A). The purpose of the estimation is
to test for the relation between the model’s independent and dependent variable given the
instrumental variables. In two-stage least-squares regression, the first stage estimates values of
the problematic predictor(s) and in the second stage, the dependent variable is regressed on
those values to obtain valid estimates of the model. Since asymptotic efficiency is not
guaranteed, it is important to apply statistical tests for model-misspecification. Estimating
Sargan’s (1964) general test of misspecification does this. Moreover, R2 is not a valid measure
of the goodness of fit in two-stage least-square models. Instead we propose that Pesaran &
19
Smith’s (1994) generalized R2 (GR2) be used. Since they are not widely used (or included in
software packages), the computation of Sargan’s test and GR2 are shown in appendix A.
Data
Data from the 1995 survey. The data in this study originate from a survey conducted
during the winter of 1995 supplemented by financial data subsequently acquired from an
independent publicly available source. The survey was conducted by three reasearchers in
association with Kommunekemi, a state-owned Danish firm that processes hazardous waste.
For most types of hazardous waste, Kommunekemi is the mandatory alternative for Danish
firms. It is important to note that all companies in our survey are customers of Kommunekemi
and that the relationship between Kommunekemi and their customers reflects a high degree of
mutual trust. For example, Kommunekemi has helped with expertise in a number of incidents,
such as explosions and chemical spills. Information about such incidents is very sensitive and,
clearly, mutual trust is important. All respondents were told that Kommunekemi would use
their questionnaire on a confidential basis and that dissemination of information to other
parties would be in anonymous form. Subsequently, we were allowed to use the questionnaire
for research purposes subject to the condition that results are reported in anonymous form only.
The population was defined as all Danish firms that produce hazardous waste as a byproduct of their primary activities (approximately 10,000 firms). Examples of firms included in
the survey are electroplating firms and producers of paint, chemicals and medical products. In
operational terms, the firms in the target population were identified on the basis of archival
information possesed by Kommunekemi. After exclusion of the smallest waste producers (less
than 5 tons per year), and state or municipality owned companies, the target population totaled
858 firms that each produces more than 5 tons of hazardous waste each year (90% of the total
volume of hazardous waste produced in Denmark). Clearly, these companies are likely to be
among the firms most influenced by stakeholder claims related to the natural environment.
A disproportionate sample with two strata was drawn. One stratum included the 36
major producers, defined as the companies in the sample which produce 50% of the total
volume of hazardous waste. The second stratum included the remaining 822 producers, 50% of
20
these producers were randomly selected and mailed a self-administered questionnaire. The
questionnaire was developed on basis of information obtained from Kommunekemi, including
a previous large-scale study conducted by Kommunekemi supplemented by theory on
environmental behaviour of firms. A pre-test of the questionnaire was then conducted among
seven firms belonging to the target population and the response used to validate and correct the
items. All the firms involved in the pre-test were excluded from participation in the survey.
The response rate was 40%. Among the 36 largest producers, 31 qualitative interviews
of one to two hours duration were conducted (the present author conducted twenty six
interviews and a research assistant conducted the remaining five). The interviews were
conducted in connection with the collection of the questionnaires. Typically, the CEO as well
as the person responsible for handling hazardous waste participated in the interview. In a few
instances, there were additional members of the management team present. As only five firms
refused to participate, a total of 31 firms were interviewed. The purpose of these qualitative
interviews was to get more detailed information, to validate the response on the questionnaire,
to probe into some sensitive areas not relevant for the present paper (accidents, chemical spills
and the like) and to elucidate whether the answers represented idiosyncratic or organizationwide response.
In the large firms, the respondents who filled out the questionnaire were typically the
person in charge of the Environment, Health and Safety (EH&S) function. In firms without an
EH&S function the respondent was typically the CEO of the company. In both instances there
is no doubt that the respondent possessed adequate knowledge. Also note that the interviews
among the large firms allowed triangulation of response. Since there was largely agreement
among the persons who participated in the interviews, we can rule out common methods bias
in the case of the large firms.
After we received the questionnaires, all non-respondents were contacted by telephone
and were asked to answer a limited number of key questions (the questionnaire items that had
the highest variance). We then conducted a non-response analysis (using t-tests and
correlations) which involved comparison of response obtained from respondents and non-
21
respondents (a summary of these analyses is available from the author). Since we only found
few significant deviations, the findings can be generalized to the population with a high degree
of confidence.
Data from the 1999 follow-up survey. The follow-up survey was conducted in 1999 on
six selected Danish industries. The industries were chosen according to the criterion that they
should experience different levels of uncertainty and all be subject to problems of information
extensiveness. The selected industries were the chemical-, medical-, paint-, electronics-,
textile- and dairy-industry. The three first industries overlap with the original 1995 survey and
the latter three were selected because they experience less uncertianty related to environmental
issues. In operational terms, the industries were identified on basis of the firm’s NACE code
for 1999 as registered in the publicly available database CD-Direct. A limit of employees >10
was used as cut-off point. This resulted in a sampling frame of 1007 firms with more than ten
employees. Based on telephone contact, a number of firms was excluded and the sampling
frame was adjusted to 908 firms. Out of these, 545 firms accepted to participate in the survey
and were mailed a self-administered questionnaire to be returned by surface mail. Nonrespondents were subsequently contacted by telephone in order to inspire response or,
alternatively, elucidate a reason for non-response. The 545 firms accepting participation were
divided into two groups: group 1 comprising 146 large or medium-sized firms with 50 or more
employees, and, group 2 comprising 399 small firms with less than 50 employees. The large or
medium-sized firms (group 1) were mailed two questionnaires: one directed to the firm’s CEO
asking for detailed information at the strategic level and another directed to the firm’s
environmental manager (or the person with equivalent responsibility) asking for detailed
information regarding the firm’s environmental practices. The small firms (group 2) were only
mailed one questionnaire, directed to the firm’s CEO, comprising a reduced form of the
combination of questionnaires. A total of 280 out of 908 firms returned at least one completed
questionnaire to yield an overall response rate of 30.8%. In view of the rather large material,
which had to be completed, and the relatively modest interest in the content of the survey on
part of at least some firms in the sample frame, this result compares very favourably with the
22
response rates reported in similar surveys.
Measuring financial performance. The financial data used to measure performance
were obtained from a publicly accessible electronic database (CD-Direct, published by
Købmandstandens Oplysningsbureau, Denmark). Financial performance is measured as the
return on assets (ROA) in the period up to and including the year in which the survey was
conducted (1991-95). It was only possible to obtain valid ROA for 141 firms since the rest of
the sample consisted of business units that were members of larger firms, and, therefore,
independent, external data were not available. Of these, one firm was excluded because of an
extremely low ROA (-425%) in the most recent year (1995). This firm is clearly an outlier and
has since the completion of the reported survey ceased to exist. In the present study, I use longterm financial performance measured as the cumulated ROA in the period 1991-95:
roai = (1+roa91/100)×(1+roa92/100)×(1+roa93/100)×(1+roa94/100)×(1+roa95/100).
It should be noted that all analyses were repeated with data on the most recent year
(1995). The results of these analyses (which can be obtained from the author) show a weaker
effect compared to the analyses using the cumulative measure but are otherwise consistent with
(same sign) the results reported in the present study.
Measuring cognitive frames. In the present study we use two broad frames (myopic
frame and strategic frame) as a proxy for cognitive frames. Furthermore, the scope is limited to
integration of issues related to the natural environment. According to this operationalization, a
strategic frame and a myopic frame related to environmental issues are broad cognitive frames
that act to direct managerial attention. As evidenced in the literature on the role of strategic
versus myopic frames, these constructs provide excellent illustration of cognitive frames since
they are commonly reported to serve as prototypes for concept driven information processing
(see e.g. Fletcher & Wright, 1996). As previously noted, the empirical literature on the firm’s
environmental (green) strategy further suggest their importance (e.g. Klassen & Whybark,
1999 and Fischer & Schot, 1993).
Both the strategic and the myopic frame was measured by two items (see Table 1,
below). The items chosen for the myopic frame were both related to costs. The reason for this
23
choice is that it is clearly problematic to directly ask managers if they are myopic with respect
to environmental issues. Very few would probably admit to be so. Since the literature on
corporate greening suggests that a myopic frame is associated with a short-term focus on costs
(Hoffman, 1999; Fischer & Schot, 1993), two cost-items were chosen for the myopic frame.
The assumption that a myopic frame was closely associated with short-term cost consideration
was supported during the qualitative interviews. The impression that a strategic and a myopic
frame were shared throughout the organization was also supported in the multi-person
interviews. In terms of observable data, we obtained information on the firms’ investment in
reducing its environmental impact (did the firm use life cycle analysis, was an environmental
management system in place, was it certified, how many employees were involved in
improving the environmental conduct of the firm etc.). An ANOVA-analysis showed that those
who had invested in any of the options we measured scored significantly lower on “myopic
frame” and higher on “strategic frame.” In sum, we have some support for the contention that
we actually measured an organization-wide strategic and myopic cognitive frame.
The assessment of scale-reliability through estimation of Cronbach’s Alpha supported
the notion that the two items in each scale capture different dimensions of a common construct
(Table 1). The myopic frame obtained an Alpha of 0.77, a sufficiently high value. By contrast,
the Alpha obtained on the construct “strategic frame” was 0.57, barely above the acceptable
limit (0.50). This value should be interpreted in view of the item-dependent nature of
Cronbach’s Alpha, which can be assessed by the general Spearman-Brown formula (Peter,
1979). Such estimation shows that the value 0.57 obtained on the two-item construct “strategic
frame” would increase to 0.80 in an equivalent five-item construct. Nevertheless, the relatively
low alpha is a source of concern.
Consequently, we included the items in a follow-up survey conducted in 1999, four
years after the orginal survey (the items were identical but in the follow-up study a seven-point
scale rather than a five-point scale was used). The follow-up survey used two different
questionnaires, one for large firms (50 employees or more) and one for small firms (less than
50 employees). For a strategic frame, we obtained an Alpha of 0.69 for large firms, 0.79 for
24
small firms and 0.77 overall (see Appendix B). For the myopic frame, the Alpha was 0.68 for
large firms, 0.71 for small firms and 0.70 overall. Both the strategic and the myopic frame are
reliable across size in the follow-up study. More important, the strategic frame has gained in
reliability over time whereas the reliability of the myopic frame has decreased sligtly. One
explanation for this observation consistent with the literature on coporate greening (see
Hoffman, 1999) is that we are observing a shift in cognitive frames so the strategic frame is
becoming increasingly stable and the myopic frame perhaps less so. A possible reason for this
shift in accordance with the hypotheses proposed in the present paper, is that the myopic frame
is simply not adaptive whereas the strategic frame is.
In sum, it can be concluded that a strategic frame has become increasingly reliable over
time whereas the myopic frame is still reliable although its reliability has slightly decreased.
This indicates that, although relatively stable, the measured cognitive frames are indeed
shifting over time. Therefore, the particular results reported in the present study cannot be
assumed to hold over time. Despite this fact, the underlying general result, that broad frames,
through information processing and uncertainty, can influence performance over relatively
long periods of time, remains valid if both the constructs uncertainty and information
processing are reliable over time. As explained in the following, this was, indeed, the case.
Further support for the reliability of the results is provided by correlations among constructs as
shown in Table 2, below.
Measuring organization-level information processing. In the present study we
measure organization-level information processing within the specific context related to
environmental issues. We used five items, which according to the literature captures
organization-level information processing (Tushman, 1978; Van de Ven et al., 1976), i.e.,
organization-wide discussion, policy formulation, distribution of knowledge about policy, top
management involvement in policy formation and employee involvement in policy
implementation. The assessment of scale-reliability through estimation of Cronbach’s Alpha
(0.87) indicate that the five items capture different dimensions of the common construct,
organization-level information processing in relation to the natural environment very well. The
25
31 qualitative interviews conducted in the present study were further used to assess face
validity of the construct used to measure information processing. Consider the following
typical excerpt from a large chemical producer as illustration, “The environmental policy
implies that we discuss and look through things regularly. ... you talk to people out there...” In
numerous instances, the 31 interviews clearly showed that the five items used in the present
study capture organization-level information processing. Finally, the follow-up study included
the measure for information processing. The overall Alpha obtained was 0.93, a value that
indicates very good stability of this construct across the four-year period we have studied as
well as high reliability across samples.
Measuring organization-level uncertainty. Uncertainty regarding environmental issues
was measured by two items and the value of Cronbach’s Alpha supported the notion that the
two items capture different dimensions of a common construct (Table 1). Since the concept of
uncertainty has many meanings, it should be noted that this construct captures the difficulties
in representing considerations related to the natural environment in terms of operations and
management. This use of the concept of uncertainty corresponds well to the notion of
Knightian or Keynesian uncertainty as situations in which point probabilities are not assigned
to outcomes of actions (Runde, 1998). As aforementioned, the item-dependent nature of
Cronbach’s Alpha has to be taken into consideration when interpreting scale-reliability.
Estimation using the general Spearman-brown formula shows that a two-item Alpha of 0.60
would increase to 0.82 in an equivalent five-item (and 0.88 in a seven-item) construct.
Also the uncertainty construct was measured in our follow-up sample. We obtained a
value of 0.80 for large firms, 0.67 for small firms and 0.73 overall. Thus, the construct
uncertainty related to environmental issues remains valid over the four-year period we have
studied and across small and large firms in the follow-up sample. As a result of the qualitative
interviews, which indicated that our measure captured two important organization-level
dimensions of uncertainty related to environmental issues, it became clear that there were at
least three further important uncertainty dimensions we had not considered. On this basis, we
developed three items that may be added to the uncertainty scale in future studies: (1)
26
uncertainty regarding costs caused by environmental legislation, (2) uncertainty regarding the
relative priority of local, regional and national demands related to environmental performance,
and (3) uncertainty regarding the relative priority of various different stakeholder claims
related to environmental issues. The Alpha obtained for the five-item scale including these
three items plus the two items used in the present study was 0.87. Apart from demonstrating
the increase in Alpha predicted by the Spearman-Brown formula, this result indicates that the
two-item measure used in the present study reasonably well captures two important dimensions
related to the overall uncertainty construct.
Controls. Measures were included to control for the firm’s size, age and the total
amount of dangerous waste produced. This is motivated by the following considerations. It is
possible that the firm’s size will have a positive influence on both performance and the general
level of information processing. By contrast, the age of the firm could influence the firm’s
performance and general level of information processing negatively due to either inertia
associated with old age or limited experience associated with newness (Hannan & Freeman,
1989). Since we measure information processing related to environmental issues, it is also
necessary to consider if the level of information processing is merely influenced by the amount
of hazardous waste produced.
The firm’s size is measured by the number of employees on basis of publicly available
archival data. The same source was used to obtain data on the firm’s age. The total amount of
dangerous waste was obtained through the questionnaire.
----------------------------------Table 1 about here
----------------------------------Stability over time. As noted in the above, the constructs information processing and
uncertainty remain stable over time. There is also support that the cognitive frames studied in
the present article remain relatively stable. The strategic frame has become increasingly
reliable over time whereas the myopic frame remains reliable. Note that the increasing
reliability of the strategic frame is in good accord with the shift in the chemical industry’s
27
treatment of environmental issues in the US (Hoffman, 1999) and the emergence of a general
awarenes of the strategic importance of such issues in the Scandinavian countries (Wolff,
1995).
----------------------------------Table 2 about here
----------------------------------To further assess reliability over time, the correlations among constructs were computed for
the 1995 sample and the two samples obtained in 1999, i.e. a separate sample of large firms
(50 employees or more) and small firms (less than 50 emplyees). The correlations for the
combined sample of large and small firms for 1999 were also computed. As can be seen from
Table 2, the correlation between information processing and uncertainty is remarkably stable,
both across samples and across time. And so is the relation between a strategic frame and both
information processing and uncertainty. So, all the differences in correlation among constructs
can be traced to the myopic frame. First, the strategic and the myopic frame are negatively
correlated for the 1995 sample but not for any of the 1999 samples. Among the large firms in
the 1999 sample, the negative correlation is reproduced. It is not significant, however. Since
the 1995 sample included producers of hazardous waste and the 1999 sample of large firms
included these, this result seems reasonable. A strategic and a myopic frame associated with
environmental issues become opposites as the firm is forced to consider such issues. Second,
there is significant negative correlation between a myopic frame and information processing in
the 1999 samples but not in the 1995 sample. As previously noted, H1.2 assumes that the
minimum level of information processing has not yet been reached by the organization. This
seems to be the case for the 1999 samples, which includes firms that are only marginally
influenced by environmental issues, but not for the 1995 sample (in effect refuting H1.2).
Third, the relation between a myopic frame and uncertainty is positive in all samples but
weaker in the 1999 samples. Again, this result makes sense since the 1999 samples included
firms that were much less influenced by environmental issues than those in the 1995 sample.
In sum, we can conclude that not only does the reliability of the contructs remain stable
28
over time but so does the relation among the constructs strategic frame, uncertainty and
information processing. The difference across time lies in the relation among a myopic frame
and the other constructs. In turn, this difference can reasonably be explained by the fact that
the 1999 sample included firms that, compared to the producers of hazardous waste included in
the 1995 sample, were only marginally influence by environmental issues.
RESULTS
The ensuing analysis follows the four-step procedure for estimation of difficult pathmodels described in the above. All analyses were conducted on the 1995 sample.
Estimation: step one. Estimation in step one (and two) involves three sets of linear
regression-equations (OLS-estimation) following Baron and Kenny’s (1986) suggestions for
estimation of mediator effects. Table 3a shows the first set of analyses. In model 1.0, a
strategic frame and a myopic frame related to environmental issues have asymmetric effects on
information processing. Supporting H1.1 but rejecting H1.2, the relation between a myopic
frame and information processing is absent whereas the relation between a strategic frame and
information processing is positive and significant. As model 1.1 shows, the controls only have
a minor effect.
Furthermore, the results show an inverse relation between a myopic and a strategic
frame regarding the uncertainty related to issues of the natural environment. In model 2.0, a
strategic frame reduces uncertainty related to environmental issues whereas a myopic frame
increases uncertainty, lending support to H1.3 and H1.4. Note that when the variable
“information processing” is included in model 3.0, the effect of the independent variable
“strategic frame” is no longer significant. These results lend strong support to the idea that
information processing mediates the effect between a strategic frame and uncertainty, thus
supporting H1.5.
First, model 1.0 shows that the effect of information processing regressed on a strategic
frame is significant and model 2.0 shows that the effect of uncertainty regressed on a strategic
frame is significant. Second, the effect of information processing is significant in model 3.0.
29
Third, the effect of a strategic frame is not significant in model 3.0. Note further, that the effect
of “myopic frame” is stable across the estimations in model 2.0 and 3.0 suggesting that the
estimated effects are robust. Inspection of model 1.1 shows that the control variable
“employees” has a positive but insignificant influence on information processing. Models 2.1
and 3.1 shows that “employees” has a significant and positive influence on uncertainty. In
other words, as the firm increases in size the uncertainty regarding environmental issues also
increases. Since the controls add to the explained variance without confounding the estimates,
we conclude that the estimates in model 1.0, 2.0 and 3.0. are robust.
In sum, the results in model 1.0 to 3.1 support the idea that the relation between a
strategic frame and uncertainty related to environmental issues is mediated by information
processing. Finally, models 1.0 to 3.1 reject H1.6. A myopic frame related to environmental
issues has a direct positive effect on uncertainty related to such issues but information
processing does not mediate the effect between a myopic frame and uncertainty. This result
suggests that a myopic frame related to environmental issues, as a general cognitive frame,
leads managers to overlook the very problems that cause uncertainty, and, to maintain a
minimum level of information processing, that, in the case it was increased, could possibly
change the organization’s state of ignorance.
----------------------------------Table 3a and 3b about here
----------------------------------Estimation: step two. According to Table 3b, the results in model 5.0 show that
uncertainty related to environmental issues has a negative effect on long-term financial
performance (supporting H2.1). Model 6.0 establishes the expected positive effect of
information processing on financial performance (supporting H2.2). Model 4.0 shows that
uncertainty has a significant negative effect on information processing. Since the effect of
uncertainty disappears in model 7.0, all three conditions are fulfilled for establishing
information processing as mediator between uncertainty related to environmental issues and
long-term financial performance (supporting H2.3).
30
As can be seen in Table 3b, including the controls in model 6.1 and 7.1 does not effect
the results. By contrast, the inclusion of controls in model 4.1 has a significant effect.
Explanatory power increases due to a significant negative age-effect and a positive size-effect
(employees) on information processing. These are spurious effects due to the multicollinearity
introduced by the positive correlation among the controls. The same effects cause the reduction
of explanatory power in model 5.1. Although the inclusion of controls leads to some
misspecification of explained variance in the models where they are included, the notable
feature is that throughout models 1.0 to 7.1, the signs of the estimated regression coefficients
are unaltered and their size is remarkably stable. That is, age and size effects are minor and do
not in any way change the conclusion of step one and two.
Estimation: step three. According to model 8.0 (shown in Table 3a), there is no direct
effect of a strategic frame or a myopic frame on financial performance. Again we can discard
the marginal increase in explanatory power in model 8.1 (see Table 3a) due to the
multicollinearity introduced by the controls. The results support the idea that cognitive frames
providing broad managerial orientation effect uncertainty through information processing. This
is not sufficient to support the case of complete mediation stated in H3.1 and H3.2. Since the
necessary condition to support H3.1 and H3.2 was established in step three, we further need
step four to provide the sufficient test for a relation between cognitive frames and financial
performance.
Estimation: step four. Model 9.1, shown below in Table 4, establishes that the effect of
a myopic frame on financial performance is negative and significant when the instrumental
variable (IV) uncertainty is included. Given the results in step three, this confirms H3.2. Model
9.0 and model 9.3 are not significant and thus rejected. Moreover, model 9.2 and 9.3 are
rejected due to misspecification (p-value for Sargan’s Chsq. < 0.05).
----------------------------------Table 4 about here
----------------------------------Model 10.0 and 10.1 establish that a strategic frame has a positive effect on financial
31
performance when the instrumental variables information processing and uncertainty,
respectively, are included. This confirms H3.1. The increase in explanatory power in model
10.2, compared with model 10.0 and 10.1, further shows that both instrumental variables
contribute to better explanation. The inclusion of controls in model 10.3 results in the
previously observed marginal increase in explanatory power without altering the sign or
changing the size of the coefficient much. Note further that neither model 10.2 (p-value for
Sargan’s Chsq = 0.966) or model 10.3 (p-value for Sargan’s Chsq = 0.723) can be rejected due
to misspecification but model 10.2 provides the better fit.
As mentioned above, there was a question whether the linear assumption of H2.2 was
supported by the data. This assumption was tested by visual inspection of the data
supplemented by polynomial regression up to polynomials of order four. The graph showed a
monotonically increasing relation between information processing and performance and a
cubic regression did not fit the data significantly better (R2=0.071) than the simple linear form
(R2=0.066). Therefore, these analyses (which can be obtained from the author), supported the
linear assumption on which H2.2 is based.
A further issue of concern is that the validity of the results may be compromised by
response bias, social desirability bias or common methods bias. Note that the issue of common
methods bias only concerns analyses where financial performance was not used as independent
variable. When financial performance was used, the dependent variable (archival data) and
independent variables (questionnaire) differed in methods. Consequently, there is no concern
of common methods bias regarding these results.
We shall first address the issue of social desirability bias, which refers to the
respondent’s tendency to choose response that reflects socially approved behaviors (see e.g.
Zerbe and Paulhaus, 1987). One way to deal with this issue is to include scales of social
desirability and subsequently correlate measures with such a scale. Although appealing in
terms of tractability there are a number of problems related to this approach (Zerbe and
Paulhaus, 1987), which motivate assessment of social desirability on the basis of the rather
unique context of our survey (referred to in the above). We shall consider problems related to
32
frankness, adequate knowledge, and response adjustment (see e.g. Nunnally and Bernstein
1994: ch. 9).
According to the stated purpose of the questionnaire and the relationship between the
respondents and Kommunekemi, conscious deviations from frank response are unlikely. There
was simply no gain or loss for the respondents to over- or underemphasize the uncertainty
related to environmental issues such as the importance of various different stakeholders, since:
(1) Kommunekemi is the mandatory waste treatment alternative, (2) there is a proven trustrelationship between the firms and Kommunekemi (evidenced in the transcripts of the 31
interviews). Therefore, we have a high degree of confidence that the results reported in the
present study do not reflect social desirability response due to deviation from frank response.
As aforementioned, the respondents were either the CEO or the executive EH&S
officer. In both instances there is no doubt that the respondent possessed adequate knowledge.
Although the possibility of response adjustment cannot be ruled out, the problem would be
most severe if the questions were not previously considered or the respondents possessed
inadequate knowledge. Since both possibilities can be ruled out, it is likely that the situational
adjustment was limited. A triangulation with the transcripts of the qualitative interviews lends
support to this assessment. The clear impression reflected in the transcripts is that there is
stability in response across the two situations. The evidence contained in the questionnaires
and the interviews largely converges. For the above reasons it is not likely that our results
reflect severe bias due to socially desirable response.
The possibility of common methods bias only concerns results where information
processing and uncertainty were used as independent variable. Since the constructs information
processing and uncertainty were reliable over time and across samples, common methods bias
can effectively ruled out for these constructs. This leaves the two constructs measuring
cognitive frames. Here, we must conclude that the strategic frame has increased in reliability
over time and the myopic frame has become slightly less reliable. This result indicates that
common methods bias is not a severe problem for the measures of cognitive frames. In sum,
common methods bias does not seem to plaque the data reported in the present paper.
33
DISCUSSION AND CONCLUSION
The present study extends the RBV by appealing to cognitive frames as codeterminants of persistent performance differentials and the empirical test supports the
suggested causal chain between cognitive frames and financial performance through
information processing and uncertainty. Furthermore, the present study introduced a novel
method to estimate path models when the usual approach is infeasible. In the following we
shall focus on the theoretical implications of our results.
The general theoretical basis underlying our argument is Simon’s (1955, 1987) work
on bounded rationality and the derived idea of imperfect environmental matching. Further and
massive empirical and theoretical support for the assumption of imperfect environmental
matching can be found in recent studies of error in decision-making (Reason, 1990). According
to Reason (1990), the imperfect correspondence between cognitive frames and social
regularities is a result of the simplification involved in elimination of differences among
occurrences perceived as similar. We base our view that cognitive frames generally direct
information processing on this simplification hypothesis, combined with the assumption of a
relatively persistent cognitive rigidity. Both assumptions have been challenged, however.
To some extent, the simplification hypothesis implies a conjecture of simple-minded
decision-makers, a view challenged by Walsh (1988). Although Walsh’s (1988) results speak
loud and clear against the hypothesis of simplification biases that direct information
processing, the evidence in favor of the alternative hypothesis seems to speak equally loud.
Walsh’s (1988) assertion that not much evidence shows that managers are actually victims of
suboptimal information processing strategies seems to pass over a number of relevant studies
too hastily. For example, Hogarth and Makridadis (1981) review 67 empirical studies on biases
in acquisition and processing of information. All 67 studies support the idea of simplification
biases. Even if we find an impressive amount of evidence supporting the idea of limits to
information processing, Walsh’s (1988) defense of managerial intelligence is well taken.
Rosman et al.’s (1994) warning that agents are portrayed as either rigid or flexible is in the
same spirit. It all depends on the particular talent of the decision-maker, the task at hand, the
34
situation, and the context of the decision.
In the present study, we have studied experienced decision-makers that processed
information associated with environmental issues. All were employees in firms that produce
large amounts of hazardous waste and as evidenced in Klassen & Whybark (1999), such issues
are a source of uncertainty for firms. We have further assumed that general cognitive frames
are rigid. The 31 interviews conducted with a subset of the respondents supported this
impression. By contrast, the results reported from the follow-up study suggested that, over a
four-year period, the strategic frame related to environmental issues has become increasingly
stable and the myopic frame slightly less so. Although decision-makers do not shift between a
myopic and a strategic frame on a daily basis, the data, in support of Rosman et al. (1994),
show that broad cognitive frames may shift gradually over time.
The theoretical interest in the present study is in the effect relatively stable cognitive
frames can have on performance differentials. Indeed, the empirical results support the general
idea that broad cognitive frames may influence financial performance through uncertainty and
information processing. There is no reason to believe that this effect is specific to
environmental issues. The general implication is that enduring cognitive frames may channel
the direction of information processing activities, influence the level of uncertainty and thus
influence performance. The results thus support the general proposition that firm-specific
cognitive frames are important co-determinants of persistent performance differentials. And
although transient, the effect can still be present over a number of years. The issue of
transience also suggests that an important question for future empirical research is whether
more specific cognitive frames are more transient than the broader frames within which they
are nested. In general, the question regarding the stability of broad cognitive frames and their
relation to performance seems a crucial one to address in future research.
35
REFERENCES
Arrow KJ. 1974. The Limits of Organization, New York: W. W. Norton & Company.
Barr PS. 1998. Adapting to Unfamiliar Environmental Events: A Look at the Evolution of
Interpretation and Its Role in Strategic Change. Organization Science 9: 644-667.
Baron RM, Kenny DA. 1986. The Moderator-Mediator Variable Distinction in Social
Psychological Research: Conceptual, Strategic, and Statistical Considerations. Journal
of Personality and Social Psychology, 51(6): 1173-1182.
Bartunek JM. 1984. Changing Interpretive Schemes and Organizational Restructuring: The
Example of Religious Order. Administrative Science Quarterly 29: 355-372.
Beyer JM. 1981. Ideologies, values, and Decision-Making in Organizations. In Handbook of
Organizational Design, Nystrom PC, Starbuck WH. (eds.). Oxford: Oxford University
Press; 166-202.
Brown JS, Collins A, Duguid P. 1989. Situated Cognition and the Culture of Learning.
Educational Researcher 18: 32-42.
Campbell DT, Stanley JC. 1966. Experimental and Quasiexperimental Designs for Research.
Boston: Houghton Mifflin.
Dess GG, Picken JC. 1999. Creating Competitive (Dis)advantage: Learning From Food Lion's
Freefall. Academy of Management Executive 13: 97-111.
Dockery K, Fawcett B. 1998. The Teams. An Oral History of the U.S. Navy Seals, New York:
William Morrow and Company, Inc.
Eden C, Spender JC. 1998. Managerial and Organizational Cognition. Theory, Methods and
Research, London: Sage Publications Ltd.
Egelhoff WG. 1982. Strategy and Structure in Multinational Corporations: An Information
Processing Approach. Administrative Science Quarterly 27: 435-458.
Fischer K, Schot J. (Eds.) 1993. Environmental Strategies for Industry. Washington D.C.:
Island Press.
Fletcher K, Wright G. 1996. The strategic context for information systems use: An empirical
study of the financial services industry. International Journal of Information
Management, 16: 119-131.
Galbraith JR. 1973. Designing Complex Organizations, Reading, MA: Addison-Wesley.
Giddens A. 1979. Central Problems in Social Theory, Berkely, CA.: University of California
Press.
Giddens A. 1984. The Constitution of Society , Cambridge: Polity Press.
Harrison JS, Freeman RE. 1999. Stakeholders, Social Responsibility, and Performance:
36
Empirical Evidence and Theoretical Perspectives, Academy of management Journal,
42(5): 479-487.
Hoffman AJ. 1999. Institutional Evolution and Change: Environmentalism and the U.S.
Chemical Industry. Academy of Management Journal 42: 351-371.
Hogarth RM, Makridakis S. 1981. Forecasting and Planning: An Evaluation. Management
Science, 27(2): 115-138.
Jelinek M, Smircich L, Hirsch PM. 1983. Introduction: A Code of Many Colors.
Administrative Science Quarterly 28: 331-338.
Kellaris JJ, Dahlstrom RF, Boyle BA. 1996. Contextual bias in ethical judgment of marketing
practices. Psychology & Marketing, 13: 677-694.
Kiesler S, Sproull L. 1982. Managerial Response to Changing Environments: Perspectives on
Problem Sensing from Social Cognition. Administrative Science Quarterly, 27:
548-570.
Klassen RD, Whybark DC. 1999. The Impact of Environmental Technologies on
Manufacturing Performance, Academy of Management Journal, 42(6): 599-615.
Kuhn TS. 1970. The Structure of Scientific Revolutions, Chicago: University of Chicago
Press.
Levinthal DA, March JG. 1993. The Myopia of Learning. Strategic Management Journal, 14:
95-112.
Luce RD, Raiffa H. 1957. Games and Decisions, London: Wiley.
March JG. 1988. Variable Risk Preferences and Adaptive Aspirations. Journal of Economic
Behavior and Organization, 9: 5-24.
March JG. 1991. Exploration and Exploitation in Organizational Learning. Organization
Science, 2(1): 71-87.
Minsky M. 1975. A Framework for Representing Knowledge. In The Psychology of Computer
Vision, Winston PH. (ed.). New York: McGraw-Hill Inc.; 240-247.
Nonaka I, Takeuchi H. 1995. The Knowledge Creating Company, Oxford: Oxford University
Press.
Nunnally JC, Bernstein IH. 1994. Psychometric Theory (3rd. ed.). New York: McGraw-Hill
Inc.
Perlow LA. 1999. The Time Famine: Toward a Sociology of Work Time. Administrative
Science Quarterly, 44: 57-81.
Pesaran MH, Smith RJ. 1994. A Generalized R2 Criterion for Regression Models Estimated by
the Instrumental Variables Method. Econometrica, 62: 705-710.
37
Peter JP. 1979. Reliability: A Review of Psychometric Basics and Recent Marketing Practices.
Journal of Marketing Research, 16: 6-17.
Peteraf MA. 1993. The Cornerstones of Competitive Advantage: A Resource-Based View,
Strategic Management Journal, 14: 179-191.
Pfeffer J. 1997. New Directions for Organization Theory. Problems and Prospects. New York
and Oxford: Oxford University Press.
Rabin M. 1998. Psychology and Economics. Journal of Economic Literature, XXXVI(1):
11-46.
Ranson S, Hinings B, Greenwood R. 1980. The Structuring of Organizational Structures.
Administrative Science Quarterly 25: 1-17.
Reason J. 1990. Human Error. Cambridge: Cambridge University Press
Rosman A, Lubatkin M, O'Neill H. 1994. Rigidity in Decision Behaviors: A Within-Subject
Test of Information Acquisition using Strategic and Financial Information Cues.
Academy of Management Journal, 37(4): 1017-1033.
Rumelt RP, Schendel DE, Teece DJ. 1995. Fundamental Issues in Strategy. A research
Agenda. Boston, MA: Harvard Business School Press.
Runde J. 1998. Clarifying Frank Knight's Discussion of the Meaning of Risk and Uncertainty.
Cambridge Journal of Economics, 22(5): 539-546.
Sargan JD. 1964. Wages and Prices in the United Kingdom: A Study in Economic
Methodology. In Econometrics Analysis for National Economic Planning, Hart PE,
Mills G, Whitaker JK. (eds.). London: Butterworth.
Schutz A. 1967. The Phenomenology of the Social World (With an introduction by George
Walsh), London: Heinemann Educational Books.
Shrivastava P. 1992. Corporate Self-Greenewal: Strategic Responses to Environmentalism.
Business Strategy and the Environment, 1: 9-21.
Shrivastava P, Schneider S. 1984. Organizational Frames of Reference. Human Relations, 37:
795-809.
Simon HA. 1955. A Behavioral Model of Rational Choice. Quarterly Journal of Economics
69: 99-118.
Simon HA. 1987. Rationality in Psychology and Economics. In Rational Choice. The Contrast
between Economics and Psychology, In Hogarth RM, Reder MW. (eds.). Chicago: The
University of Chicago Press; 25-40.
Starbuck WH. 1982. Congealing Oil: Inventing Ideologies to Justify Acting Ideologies Out.
Journal of Management Studies 19: 3-27.
Tahara I. 1994. Cognitive Information Processing. Tokyo: Ohmsha, Ltd.
38
Teece DJ, Pisano G, Shuen A. 1997. Dynamic Capabilities and Strategic Management,
Strategic Management Journal, 18: 509-535.
Thomas AB, Al-Maskati H. 1997. Contextual influences on thinking in organizations: Learner
and tutor orientations to organizational learning. Journal of Management Studies, 34:
851-870.
Tsoukas H. 1996. The Firm as a Distributed Knowledge System: A Constructionist Approach.
Strategic Management Journal 17 (Special Issue, Winter): 11-26.
Tushman ML, Romanelli E. 1985. Organizational Evolution: A Metamorphosis Model of
Convergence and Reorientation. In: Research in Organizational Behavior, Vol. 7,
Cummings LL, Staw B. (eds.). Greenwich, CT: JAI Press; 171-222.
Tushman ML, Nadler DA. 1978. Information Processing as an Integrating Concept in
Organization Design. Academy of Management Review 3: 613-624.
Tversky A, Kahneman D. 1981. The Framing of Decisions and the Psychology of Choice.
Science 211: 453-458.
Van de Ven AH, Delbecq AL, Koenig R. Jr. 1976. Determinants of Coordination Modes
Within Organizations. American Sociological Review 41: 322-338.
Van Maanen J. 1973. Observations on the Making of Policemen. Human Organization 32:
407-418.
Walsh JP. 1988. Selectivity and Selective Perception: An Investigation of Managers' Belief
Structures and Information Processing. Academy of Management Journal, 31(4):
873-896.
Weick K, Bougnon MG. 1986. Organizations as Cognitive Maps. Charting Ways to Success
and Failure. In: The Thinking Organization: Dynamics of Social Cognition, Sims HP,
Gioia DA. (eds.). San Francisco: Jossey-Bass; 102-135.
Wolff R. 1995. The Nordic Business Environmental Barometer. Oslo: Gothenburg Research
Institute and Bedriftsøkonomens Forlag A/S.
Zerbe WJ, Paulhaus DL. 1987. Socially Desirable Responding in Organizational Behavior: A
Reconceptualization. Academy of Management Review, 12: 250-264.
39
APPENDIX A
Two-stage least-squares
The first step in two stage least squares estimation involves computation of the fitted values of
the independent observations X on the n×s matrix Z of instrumental variables (n observations
on s instrumental variables and sk) by means of a n×n projection matrix Pz. As shown in the
two equations below, the second step then involves computing the IV-coefficients by
regressing y on X-hat. Estimations in the two steps are obtained through OLS-regression.
Test statistics in two-stage least-squares
The appropriate measure for assessing the goodness of fit in models involving instrumental
variables (such as the linear two-stage least-squares method) is the generalized R2 (GR2)
X = P z X
y = β IV X + e 2 S L S
proposed by Pesaran & Smith (1994). It is computed by the following formula where e2SLS are
the residuals from step two in the two-stage least-squares estimation (as shown in the equations
in the above). The first formula shows how the GR2 is computed and the second formula shows
how this measure is adjusted for degrees of freedom.
GR2 = 1−
( e ′ 2 S L Se 2 S L S )


 ∑ ( yt − y ) 2 
 t =1

n
 n − 1 
G R 2 = 1 − 
(1 − G R 2 )
n − k
Misspecification test of two-stage least-squares models
It is well known that regression involving instrumental variables is problematic due to the
possibility of misspecification. The test proposed by Sargan (1964) provides the necessary
statististics to evaluate if a model is misspecified. The test is based on the following statistics
given the null hypothesis that the s (s>k) instrumental variables contained in Z are valid
instruments:
Under the null-hypothesis, Sargan misspecification (SM) statistics is asymptotically 32
distributed with s-k degrees of freedom (s is the number of instrumental variables and k is the
number of ’s estimated).
χ 2 SM =
Q ( βIV )
σ 2 IV
′
Q ( βIV ) = (y − X βIV ) P z (y − X βIV )
40
APPENDIX B
DATA FROM THE FOLLOW-UP STUDY, CONDUCTED 1999
Means, Standard Deviations and Correlations
mean
Myopic frame related to environmental issues
Environmental intiatives make products costly
Environmental initiatives decrease competitiveness
4.84
3.81
s.d.
1.70
1.87
.54*
Scale: (1) Disagree completely, -- (7) Agree completely. Cronbachs Alpha for additive scale: Large (0.68), Small (0.71), Overall (0.70).
Strategic frame related to environmental issues
Environmental awareness is a strategic advantage
Environmental certification gives competitive
advantage
4.14
4.86
1.87
1.59
.66***
Scale: (1) Disagree completely, -- (7) Agree completely. Cronbachs Alpha for additive scale: Large (0.69), Small (0.79), Overall (0.77).
Degree of Information Processing Related to the Natural Environment
We often discuss env. problems in our firm
3.99 1.89
Our firm has formulated a clear env. policy
3.75 2.33 .63***
All employees know our environmental policy
3.63 2.20 .58**
Top management is strongly involved with the
4.08 2.17 .72***
firm's environmental policy
All employees take actively part in the
3.49 1.93 .67***
implementation of the firm's env. policy
.81***
.76***
.74***
.77***
.82***
.79***
Scale: (1) Disagree completely, -- (7) Agree completely. Cronbachs Alpha for additive scale: Large (0.91), Small (0.92), Overall (0.93).
Uncertainty related to environmental issues
Uncertainty leads to delays in handling
environmental problems
We are generally uncertain how to integrate
environmental issues in management and
operations
3.30
1.77
3.50
1.82 .58***
Scale: (1) Disagree completely, -- (7) Agree completely. Cronbachs Alpha for additive scale: Large (0.80), Small (0.67), Overall (0.73).
† <0.10, *p<0.05, **p<0.01, ***p<0.001
41
TABLE 1
Means, Standard Deviations and Correlations
Economic Performance
ROA 1995
ROA 1994
ROA 1993
ROA 1992
ROA 1991
Cumulative ROA 1991-95
Myopic frame related to environmental issues
Environmental intiatives make products costly
Environmental initiatives decrease competitiveness
mean
1.08
1.08
1.07
1.07
1.07
1.46
3.46
2.74
s.d. 1
2
3
4
5
.09
.09
.73***
.08
.59*** .67***
.08
.47*** .46*** .62***
.07
.37*** .43*** .52*** .64***
.47
.79*** .83*** .84*** .77*** .72***
1.21
1.30 .63***
Scale: (1) Disagree completely ,-- (5) Agree completely. Cronbachs Alpha for additive scale: .77
Strategic frame related to environmental issues
Environmental awareness is a strategic advantage
Environmental certification gives competitive
advantage
3.46
4.14
1.29
.87
.43***
Scale: (1) Disagree completely ,-- (5) Agree completely. Cronbachs Alpha for additive scale: .57
Degree of Information Processing Related to the Natural Environment
We often discuss env. problems in our firm
4.08 1.00
Our firm has formulated a clear env. policy
3.75 1.24 .44***
All employees know our environmental policy
3.45 1.19 .48***
Top management is strongly involved with the
3.92 1.18 .48***
firm's environmental policy
All employees take actively part in the
3.46 1.10 .36***
implementation of the firm's env. policy
.73***
.73***
.64***
.56***
.56***
.66***
Scale: (1) Disagree completely ,-- (5) Agree completely. Cronbachs Alpha for additive scale: .87
Uncertainty related to environmental issues
Uncertainty leads to delays in handling
environmental problems
We are generally uncertain how to integrate
environmental issues in management and
operations
2.02
1.05
2.03
1.10 .44***
Scale: (1) Disagree completely ,-- (5) Agree completely. Cronbachs Alpha for additive scale: .60
Controls
Age of the firm (years)
52
39
Number of employees
471 1451 0.174*
Total amount of dangerous waste (tonnes)
156
379 0.091
0.172†
† <0.10, *p<0.05, **p<0.01, ***p<0.001.
42
TABLE 2
Correlations Among Constructs, 1995 and 1999
1995 Sample
Producers of hazardous waste
Strategic Frame
Myopic Frame
-0,16*
Information Processing 0,44*** -0,01
Uncertainty
-0,28*** 0,26*** -0,37***
1999 Sample
Large Firms
Strategic Frame
Myopic Frame
-0,16
Information Processing 0,44*** -0,14
Uncertainty
-0,27† 0,16
-0,37**
Small Firms
Strategic Frame
Myopic Frame
0,00
Information Processing 0,43*** -0,23*
Uncertainty
-0,26** 0,14
-0,39***
Combined Sample: Large and Small Firms
Strategic Frame
Myopic Frame
-0,01
Information Processing 0,38*** -0,18*
Uncertainty
-0,26*** 0,13† -0,46***
† <0.10, *p<0.05, **p<0.01, ***p<0.001
43
TABLE 3A
Regression Results
Variable
(Constant)
Model 1.0 Model 1.1 Model 2.0 Model 2.1 Model 3.0 Model 3.1 Model 8.0 Model 8.1
(0,400)
(0,459)
(0,423)
-0.484
-0.453
(0.512)
(0.226)
(0,244)
0,248**
-0.068
-0,264**
(0,084)
0,275**
-0.075
-0,309***
-0.086
-0,292**
-0.095
0,235**
-0.067
-0.109
-0.096
-0,285**
(0,099)
0,258**
-0.073
-0,156
(0,098)
0,034
(0,36)
0,115
(0,045)
0,097
(0,037)
0,161†
(0,043)
0.143
-0.066
-0.002
0,183*
0
-0.088
0
0.183
0.194
-0,094
-0.002
0,215*
(0,000)
-0,096
0
0.229
0.000
-0,060
(0,001)
-0,054
(0,000)
0,076
(0,000)
0,001
Info. Proc.
Myopic frame
Strat. frame
-0.022
-0.079
0,494***
-0.064
-0.007
-0.07
0,509***
(0,081)
0,237
-0.087
-0.002
0.112
0
0.062
(0,000)
0,267
Controls
Age
Employees
Waste
R-Square
Dependent model 1.0-1.1: Degree of information processing. Dependent model 2.0-3.1: Uncertainty
Standard errors are in parenthesis, Adjusted R-square and standardized coefficients reported. † <0.10, *p<0.05, **p<0.01, ***p<0.001
TABLE 3B
Regression Results
Variable
(Constant)
Model 4.0 Model 4.1 Model 5.0 Model 5.1 Model 6.0 Model 6.1 Model 7.0 Model 7.1
(0,324)
(0,230)
(0,097)
(0,113)
Info. Proc.
Uncertainty
-0,381*** -0,397***
(0,084)
(0,087)
Controls
Age
Employees
Waste
R-Square
0.138
-0,190*
(0,002)
0,205*
(0,000)
-0,037
0
0.172
-0,223*
(0,043)
-0,199*
(0,043)
0,042
-0,087
(0,001)
-0,045
(0,000)
-0,007
0
0,016
(0,167)
(0,173)
(0,235)
(0,247)
0,257**
(0,043)
0,281**
(0,042)
0,213*
(0,047)
-0,13
(0,047)
0,262*
(0,047)
-0,077
(0,046)
0,058
-0,04
(0,001)
-0,112
(0,000)
-0,053
0
0,058
0.068
-0,023
(0,001)
-0,103
(0,000)
0,016
(0,000)
0.055
Dependent model 4.0-4.1: Degree of information processing. Dependent model 5.0-7.1: Economic performance, long-term
Standard errors are in parenthesis, Adjusted R-square and standardized coefficients reported. † <0.10, *p<0.05, **p<0.01, ***p<0.001
44
TABLE 4
IV-Regression Results (two-stage least-squares)
Variable
(Constant)
Myopic
frame
Model
9.0
(1,777)
Model
9.1
(0,461)
Model
9.2
(0,464)
Model
9.3
(0,224)
-0,754
-0,743*
-0,734†
-0,130
(0,539)
(0,149)
(0,150)
(0,072)
Strat. Frame
IV
Info. Proc.
Uncertainty
Age
Employees
Waste
R-Square
GR-Square
X
0,003
0,069
X
X
X
0,023
0,029
0,022
0,032
Model 10.0 Model 10.1 Model 10.2 Model 10.3
(0,377)
(0,625)
(0,365)
(0,315)
0,527*
(0,099)
0,705*
(0,163)
0,550**
(0,095)
0,519**
(0,082)
X
X
X
0,029
0,029
0,057
0,076
X
X
X
X
X
0,062
0,082
X
X
X
X
X
0,000
0,003
X
0,045
0,069
Dependent model 9.0: Economic performance, long-term, p-value for model: 0,245
Dependent model 9.1: Economic performance, long-term, p-value for model: 0,049
Dependent model 9.2: Economic performance, long-term, p-value for model: 0,052; Sargan's CHSQ(1df): 3,885 (p=0,049)
Dependent model 9.3: Economic performance, long-term, p-value for model: 0,508; Sargan's CHSQ(4df): 12,016 (p=0,017)
Dependent model 10.0: Economic performance, long-term, p-value for model: 0,032
Dependent model 10.1: Economic performance, long-term, p-value for model: 0,010
Dependent model 10.2: Economic performance, long-term, p-value for model: 0,005; Sargan's CHSQ(1df): 0,002 (p=0,966)
Dependent model 10.3: Economic performance, long-term, p-value for model: 0,004; Sargan's CHSQ(4df): 2,067 (p=0,723)
Dependent model 11.1: Economic performance, long-term, p-value for model: 0,015; Sargan's CHSQ(3df): 1,069 (p=0,.785)
Standard errors are in parenthesis.
† <0.10, *p<0.05, **p<0.01, ***p<0.001
45
FIGURE 1:
Paths and hypothesized relations.
0
Strategic
Frame
+
Information
Processing
+
-
Myopic
Frame
Financial
Performance
-
-
-
0
Uncertainty
+
0
a
Since the model cannot be identified, the estimation follows the four-step “diamond-procedure” outlined below.
46
FIGURE 2
Estimating an over-identified model by decomposition into four estimation steps.
Path analysis by means of combined mediator-test (I, II)
and 2-stage least squares (III)
Z1
I
Z2
Z2
Y
II
Z1
X
X
Z1
Y
III X
Y
Z2
Z1
IV X
Y
Z2
47