Modeling new-firm growth and survival with panel data using event magnitude regression

Abstract

We introduce a new model to address three methodological biases in research on new venture growth and survival. The model offers entrepreneurship scholars numerous benefits. The biases are identified using a systematic review of 96 papers using longitudinal data published over a period of 20 years. They are: (1) distributional properties of new ventures; (2) selection bias; and (3) causal asymmetry. The biases make the popular use of normal distribution models problematic. As a potential solution, we introduce and test an event magnitude regression model approach (EMM). In this two-stage model, the first model explores the probability of four events: a firm staying the same size, expanding, contracting, or exiting. In the second stage, if the firm contracts or expands, we estimate the magnitude of the change. A suggested benefit is that researchers can better separate the likelihood of an event from its magnitude, thereby opening new avenues for research. We provide an overview of our model analyzing an example data set involving longitudinal venture level data. We provide a new package for the statistical software R. Our findings show that EMM outperforms the widely adopted normal distribution model. We discuss the benefits and consequences of our model, identify areas for future research, and offer recommendations for research practice.