Causality: Endogeneity Biases and Possible Remedies
Many, if not all, studies in accounting and information systems address causal research questions. A key feature of such questions is that they seek to establish whether a variation in X (the treatment) leads to a state change in Y (the effect). These stu
- PDF / 449,495 Bytes
- 36 Pages / 439.37 x 666.142 pts Page_size
- 22 Downloads / 183 Views
Causality: Endogeneity Biases and Possible Remedies
Many, if not all, studies in accounting and information systems address causal research questions. A key feature of such questions is that they seek to establish whether a variation in X (the treatment) leads to a state change in Y (the effect). These studies go beyond an association between two phenomena (i.e., a correlation between variables in the empirical model) to find a true cause-effect relationship. Moving from a simple association to a causal claims requires meeting a number of conditions. Consider the relationship between board independence and firm performance. A recurrent question concerns whether increasing board independence (cause) improves decision-making or firm performance (effect). Addressing this question involves several methodological issues that, if ignored, hamper the ability to make conclusive claims about the cause-effect relationship. In order to address this question, the research design should take into account two issues: Firm performance or expected performance may affect a board’s choice in appointing more or less independent directors (an example of reverse causality), and the number of independent directors varies with a number of other variables that jointly affect the main predictor and the effect of interest (an example of omitted correlated variables). So far, we have introduced several statistical methods with which to investigate relationships between variables. This chapter examines the conditions an empirical study has to meet in order to support causal claims. In so doing, the chapter compares an ideal state (e.g., a randomized experiment) with non-experimental data and discusses the issues researchers encounter in dealing with observational data. Then it offers a hands-on approach to a series of remedies to overcome the potential shortcomings in designing or executing research that seeks to make causal claims.
# Springer International Publishing Switzerland 2017 W. Mertens et al., Quantitative Data Analysis, DOI 10.1007/978-3-319-42700-3_7
99
100
7.1
7
Causality: Endogeneity Biases and Possible Remedies
Your Research Question Is Causal: What Does that Mean?
Imagine you have collected data that shows an increase in a technology’s perceived usefulness and intent to use the technology. It would be tempting to state that perceived usefulness caused that increase in intent to use (and that is often how scholars report such results). However, there are at least three reasons that such a conclusion is probably premature: First, that two things occur together or behave similarly (i.e., co-vary) is not sufficient reason to conclude that one variable causes another. They may have co-varied by chance or because of a third variable that influenced both. Second, a causal conclusion like this is allowed only when cases (people) are randomly assigned to different levels of the independent variable (perceived usefulness), when that independent variable is manipulated in a controlled environment (e.g., by assigning them to tec
Data Loading...
