The performance of entertainment products depends on their ability to transport the audience into the narrative. Deriving entertainment features (imagery and display of emotions) from screenplays, we empirically test the theory-inspired conceptual framework. The proposed framework details the nonlinear effects of imagery and display of emotions and their synergistic effects on box office performance. We correct for the sample selection bias using a sample of unproduced and produced screenplays as well as for the endogeneity due to the omitted variable bias. The results show that the increase in the relative number of characters and dialogs, decrease in scene pace, and increases in emotionality and positivity improve the box office return on investment (ROI) at a decreasing rate. A 1% desired change in entertainment features corresponds to a 1.97% increase in ROI, which is equivalent to a $1.24 million increase in revenue (assuming an average budget). Overall, this study enables film studios to identify and craft screenplays that have the potential to perform well in the box office.
Eliashberg et al. (2007, 2014) do not call these screenplay characteristics “imagery.” However, based on our conversations with a few movie directors/producers, imagery is referred to by many as the visual screenplay characteristics, such as the features captured in their study.
We also asked two independent professionals with script reading experience and training in film production to read each script and measure the completeness of emotional features captured from LIWC. The interrater reliability is 86%, suggesting a high level of agreement between the raters on the acceptability of the emotional measures.
One may argue that the domestic financial information is an incomplete measure because film studios often generate significant revenue in foreign markets. In Web Appendix B, we perform additional analyses for worldwide box office revenue and ROI. We find consistent effects of imagery features and display of emotions of screenplay on worldwide performance measures. These results are not surprising because the correlation between domestic and worldwide box office performances is very high (ρ = .95) in our sample, which suggests that movies that perform well in the U.S. also perform well globally.
VIF of the screenplay features are less than 1.3 for both revenue and ROI models. We also provide stepwise analyses in Web Appendix C and find that the coefficients are relatively stable, confirming that multicollinearity is not a major issue.
We test for the validity of the IVs with the Sargan-Hansen test (Kennedy, 2003). The null hypothesis is that the instruments are exogenous and uncorrelated with the error term, and we failed to reject the null. Thus, the instruments in the estimation are valid.
Since the IMR from the Heckman process and the endogeneity correction terms from the control function approach included in Eq. 1 are not from the original data, we use bootstrapping to compute the standard errors (Petrin & Train, 2010). The reported standard errors are derived from 1000 bootstrapped samples.