Political Networks of Companies  AOM Conference Paper
Notes on CONCLUSION
Model Performance.
The model presented provides a good explanation of the political activity response. For comparison purposes we summarize some model diagnostics for 2 trivial models, the "All Predictor" model, and the Reduced models. The reduced models are clearly preferable, as evidenced by the overall model pvalue and the percent of significant predictor coefficients. Additionally, but what is not shown in the table, is the intercept coefficient for the reduced model, which equals 43 with a 14% significance, supports the interpretation discussed above. The nonreduced intercepts are all on the order of +55. The trivial models' intercept are all significant, but the All Predictor intercept is insignificant with pvalue = 0.60.
The final model included at least one variable from each hypothesis; for those responses which the blind reduced model eliminated all variables from a hypothesis, an appropriate variable was forced into the set. These were obtained by another stepwise selection from the hypothesis variable set, and selecting only those variables which were significant on this subsequent fit. This typically added in two desired variables, which were not usually significant, but did help reduce the overall RSE, and which improved residual normality for response Q20c, as well as providing a more consistent analytical framework for the research question. The table below focuses on Q11c since that model had the highest RSE.
Model Diagnostic Comparison 

Private Meetings with Senators, Q11c 












Model 
Predictors 
% Signif. 
Model 
K 
R^{2} 
RSE 
Pvalue 
Significant 
Predictors 
Rev02 
1 
0.17 
70 
0.0004 
Rev02 
100% 







Rev02 
2 
0.17 
70 
0.002 
Rev02 
50% 
AllGive02 













AllPredictors 
26 
0.67 
64 
0.04 
Q2 Rev02, 72 
20% 





73, 83, 26 








Reduced 
8 
0.47 
58 
1.00E06 
62, 61, rev02 
75% 





72, 73, 44 








Forced Reduced 
10 
0.49 
58 
2.21E05 
Q2, Rev02, Q26, Q72, Q73, Q61, AllGive02, Q78, Q44, Q83 
50% 
Intercepts. We would generally expect the intercepts on all the response variables to be less than or equal to zero or some very small number, since access is generally a scare commodity. This is borne out except for private meetings with MOHR/staff (Q20c), which has an intercept of around +17 to +32, depending on which regression software one uses. However, in comparing standard errors, for Q20c we see that with sample size of 58 the s_{a}=31 versus that for Q8, s_{a}= 8.7, giving a 95% confidence interval for intercept a as [44, 80], versus that for Q8 of [37, 2].
NOTE: The following discussion must be compared to the new FORCED variables as well as Rehbein’s H3, not some verbiage is useful.
Excluded Variables. From a total of 29 explanatory variables, 14 were included against various responses. It is instructive to consider which variables were EXCLUDED. The Following variables were eliminated altogether (with discussion):
Q67 # senior managers worked for any level of US government.
It is surprising that those senior managers who worked for the
Largess02 100*AllGive02/(Income02*1,000,000).
It appears that the relative giving Largess is not as significant as the actual giving levels as provided in AllGive02.
Q34 Company served on an advisory committee of a regulatory agency.
Q74 Company been invited to join any of the business advisory groups to the federal
We conclude from these exclusions that involvement in the Executive branch is not as significant as directed efforts at the Legislative. (See Q67, Managers who Worked for US Government).
Q63 CEO testified at a regulatory hearing
Q64 CEO argued in industry's trade association
Usually testimony is for the record and to satisfy political imperative, not to improve political networks of witnesses. The exclusion of industry trade association makes sense because of interest is the involvement in the Congressional interests.
Q39 Coordinate with other firms in industry
This coordination activity, although helpful for competitor relationships, represents coordination outside the elected political sphere, and was excluded from the model predicting political networks.
Q81 How often does your company conduct political education programs?
Q82 How often does your company use public relations advertising in the media?
Q84 How often does your company use grassroots political programs?
These approaches are broadbased (with the expected 2% return factor). The only constituiencybuilding activity of this sort which influces the response is frequency of press conferences on public policy; perhaps the politicos pay more attention to press releases rather than more grassroots activity.
Correlations.
High correlation between Largess and number of top managers who previously worked for US Government indicates can be attributed to the experience factor.