# No other coefficients are statistically significant at 95% confidence, except NUMINST, which is

statistically significant in a direction opposite to what negotiation theory would predict.

# In view of the lack of statistical significance among controls, I now run a parsimonious

model that includes only a subset of the controls that were included in Table 3. Reducing the

number of controls is particularly useful in a small sample in order to maximize degrees of

freedom. Specifically, I eliminate the three variables under the heading of “Target Shareholder

# Profile,” none of which were statistically significant in the Table 3 regressions. I run the

stripped-down model only using the final premium as the dependent variable, over the average

trading price for 60 days prior to deal announcement. Results are similar when I run the model

on the final premium over the average trading price 30 days prior to deal announcement. The

results from this analysis are reported in Tables 4A and 4B.

[insert Tables 4A and 4B here]

Model #1 in Tables 4A & 4B shows the baseline specification, that is, the same model as

reported in Table 3 but with the target shareholder profile variables omitted. The results from

this baseline model are consistent with the results reported in Table 3, including the magnitude

and statistical significant of the TENDER coefficient. The remaining models report results from

three alternative specifications.

Models #2 and #3 focus on the influence of deal size on the results reported thus far. Freeze-

outs overall are relatively small deals (e.g., median deal size of $17.0 million among post-

# Siliconix freeze-outs, as reported in Table 1) and so while the negative coefficients for the tender

offer variables reported in Table 3 are generally large in magnitude, the differences in outcomes

between the two transactional forms may not be economically meaningful. Results reported thus

far control for deal size by including the LNVAL variable in all specifications. I now examine

27