The methodological approach follows the process of structuration6 applied by Weber et al (2013) and developed from Weick’s (1995) work on sense-making within organizations. The structuration process states institutionalization occurs over time as distinct practices become interconnected through repetition and form coherent
institutional frameworks. Applied to Figure 5.1, the theory predicts combinations of specific actions form organizational strategies in an established field subject to institutional pressure. As an established field, the theoretical assumption for the
6Structuration comes from the work of Anthony Giddens (1984) and is a recursive conceptualization of
pharmaceutical industry is that institutionalization has already occurred and, therefore, outcomes from general practices can identify the latent connections between actions.
The dataset contains a large number of items and it is unlikely that all of them operationalize a latent institutional framework; both the practice-as-strategy and structuration argument hold that not every practice is a component of an institutional framework. Applying a multi-step analysis is a method to achieve parsimony in the model and determine what measures contribute to the formation of organizational subgroups in the pharmaceutical industry. Exploratory factor analysis is the first step to identify existing connections between organizational actions. An exploratory factor analysis will address the basic question of whether latent constructs exist within the data as well as the specific measurement items that are connected. Structural equation
modeling is the second step to assess if the identified factors interconnect through a larger relational structure. The structural equation model will determine if the measures form a coherent framework and if there are causal relationships between the latent constructs.
A logical method for evaluating organizational strategy is to study the decision- making processes behind an organization’s central objective, the process known as strategy-as-practice (Schraven et al 2015, Vaara and Whittington 2012). For
pharmaceutical companies practice is developing and selling pharmaceutical products. The application of drug approval data a measure of organizational strategy is a novel contribution of this project but supported by prior research showing that these decisions are strategic actions. The advent of discovery by design research processes and studies on the pharmaceutical pipeline (Fisher et al. 2015, Sowlay and Lloyd 2010) indicate that drugs reaching the market are the result of deliberate choices made within the
organization. While using only approval data leaves out research projects that never reach the clinical testing phase or are abandoned prior to approval, this constraint presents an accurate measure of the outcomes that become subject to negotiation between
stakeholders within the field.
The first purpose of the exploratory factor analysis is to examine the dataset to determine if there are latent constructs to warrant further analysis. This is a conceptual quantitative first step in structuration (Weber 2013) because it identifies connections between different actions. The variables in the dataset represent different organizational decisions; therefore, that factors identified in the model should provide relational information on general organizational strategies in the field. Following a model fitting strategy, as opposed to model testing, allows some measure to not load on any factors without contradicting the overall theoretical expectation that latent constructs exist.
Based on the existing research, several predications can be made about possible latent constructs. Corporate financial information can reveal the organizational structure of a pharmaceutical firm and, therefore, can illustrate the focus of organizational strategy (Chandler 2005, Davis et al 2004, Roy 1997, Richard et al. 2009, Powell and Sandholtz 2012). The financial information, including the structural components of subsidiary, merger, and acquisition, form the basis of attention for the organization. A factor loading return on assets (ROA), net earnings per basic share, and marketing and administrative expenses would indicate strategies of profitability. If research and development, acquired R&D, and joint R&D cost, load together it could indicate an organizational focus on innovation.
The chemical type and therapeutic class variables illustrate organizational research strategy because each submission is a specific organizational action. The drug information is more challenging for predicting possible factor outcomes, but drawing on Fisher et al.’s (2015) work on the drug pipeline, new molecular entity application is expected to load with antineoplastic agents (cancer drugs) because cancer dominates the majority of the drugs under development. Antivirals (anti-infectives) and painkillers (central nervous system agents) are the other top drug classes in development expected to load with new molecular entity application, along with priority review and orphan drug status. New combination, new indication, and new formulation are predicted to load together since they are aspects of drug expansion and possible indications for an organizational strategy of medicalization.
In summation, the factor analysis will reveal latent constructs at the field level. These constructs are not all expected to contribute to organizational differences. It is likely some of the constructs will be the result of structural constraints placed on the organizations within the field. The contribution of the factor analysis in this research is to determine if latent constructs not readily explainable by external constraints exist. These factors are the most likely to provide the measures for the components that comprise an institutional framework.
Structural equation modeling allows for the determination of a causal relationship between the factors derived from the factor analysis (Acock 2013, Bollen 2011). This is an important next step because the central argument, field level constructs effect
organizational strategy, is unsupported by factor analysis alone since the direction of effects is unclear in first-generation statistical techniques (Bagozzi and Yi 2012). As a
second-generation statistical technique, structural equation modeling provides a method for distinguishing latent constructs that are causal indicators of organizational strategy. Furthermore, since it is theoretically possible not all of the identified factors from the factor analysis will connect in a single coherent framework, applying a model fitting strategy allows for the elimination of unrelated factors. The contribution of the structural equation model to this research is that it will identify the presence of an interconnecting framework of pharmaceutical practices. Drawing from the strategy-as-practice research, the structural equation model will indicate distinct organizational level strategies within the field.
To determine if the measure of the identified strategies create organizational subgroups, the indicators from the best fit structural equation model will be used to conduct a latent class analysis. Mo Wang and Paul Hanges (2011) proposed latent class modeling as a more robust analysis technique for identifying organizational heterogeneity compared to current methodologies relying on categorization through qualitative
techniques or stratification by demographic characteristics. Clustering analyses based on categorical variables is common practice but this method is problematic for evaluating causal claims because while differences may emerge between the groups, the researcher is unable to evaluate if the differences are a result of the characteristic used to define the groups (Wang and Hanges 2011). Latent class models can support causal claims because the techniques statistically derive the groups from a range of indicators; therefore, the researcher is able to determine which indictors are important for class separation by evaluating the class prevalence rates. The interpretation of latent classes is based on the
results of the analysis and it is inappropriate to decide a priori how many classes are in the data or how to categorize those classes (Collins and Lanza 2010, Masyn 2013).
Latent class analysis uses categorical variables, which require the transformation of the chemical type and therapeutic class variables from counts into ordinal categories. With a sample size of only 59 cases, a simple dichotomous transformation (1 = received an approval, 0 = did not receive an approval) is appropriate given the prevalence of no approvals in all categories. Using dummy variables does prevent the measurement of potential differences due to variations in approval rates, but given the small sample size, it is unlikely much variation would have be picked up in a more detailed categorical transformation.