# Interpretation of Model When Intentionally Excluding a Control Variable!

I am looking at factors affecting firm value (the dependent variable). I fitted the following model using OLS:

$log(FirmValue_i)=\beta_0 + \beta_1Divers_i+ \beta_2HQLoc_i + \epsilon_i$

where $Divers$ is the diversification discount and $HQLoc$ is the location of corporate headquarter. I also controlled for R&D expenditure, a proxy measure of information asymmetry, in my model. When I exclude R&D, the coefficient for diversification discount is 7.5% and the coefficient for NYC is 5%. When I include R&D the coefficient on diversification discount drops to 2% and the coefficient of NYC drops to 1%.

I know this means that it is actually R&D that affects firm value rather than diversification but is it correct for me to say, "R&D as proxy for information asymmetry clearly has an important effect on the dependent variable excess value. However, in the models in which we don’t control for R&D and thus for information asymmetry, the effect of location is more pronounced. This confirms the hypothesis that when there is higher information asymmetry, such as in diversified firms with complex structures, the location is more important?"

• I'm confused over your use of percentages for your coefficients. Are these expressed as marginal effects? What kind of regression model did you run? OLS? Logit? – Marquis de Carabas Aug 15 '15 at 22:14
• I'm sorry. My dependent variable is log - that's why I interpret the coefficients as %. I did an OLS regression! – Gloria Little Aug 15 '15 at 22:17
• Thanks so much for editing! I also added some control variables in the regression (e.g. profitability, leverage, etc.) but this is fine to understand the main idea! – Gloria Little Aug 15 '15 at 22:27
• Is $\beta_{RandD}$ statistically significant? How does the significance of the other coefficients change when you add $RandD$ into the model? – Marquis de Carabas Aug 15 '15 at 22:29
• The beta of R&D is always significant, so it is definitely an important control variable, but it greatly reduces my sample size (by 50%) - that's why I always run one regression with R&D & one without. I only have about 1000 diversified firms in my sample, so reducing it by 50% is not good. The significance level of the multi-segment indicator (= 1 if the firm has more than one segment, 0 otherwise) to determine the diversification discount remains significant at 1%. Unfortunately my location variables (NYC) are never significant. I still interpret them though, since its the topic of my paper – Gloria Little Aug 15 '15 at 22:34