Political Networks of Companies - AOM Conference Paper
                                                Notes on the Data

Total companies canvassad for survey:                 210
Total companies responding to survey with                  
      useable data for AOM response variables:         81
Typical size of multivariate regression:               52

Dim(Aom):  81 observations and 31 variables

Missing Values - This includes the Regulatory Uncertainty variables Q25 and Q26.
-------------------------------------------------------------------------------
Fortunately, the sample size was small enough to afford visual enumeration.  If a company did not provide survey data for the response variables (Q8, Q17, Q11c, and Q20c) then it was deleted.  

Altogether there were 47 missing values in the 81 observations, resulting in a missing data rate of 1.9% (out of 2511 total data).  Fortunately, some of these were in the same observations.  Depending on the response, 2 companies had to be dropped due to missing response data.  Each predictor variable class had only a moderate amount of missing data; these are:

        All predictors:  29 observations deleted due to missing values
        Controls:        12 observations deleted due to missing values
        Incentives:      20 observations deleted due to missing values
        Information:      9 observations deleted due to missing values
        Constituiency:    2 observations deleted due to missing values
        Uncertainty:      1 observations deleted due to missing values

The subsets of missing data combine to an overall deletion of 29 observations on the full regression, resulting in 25 degress of freedom (but with overall p-value for the model of less than 0.002).  

Several applicable companies (14) were missing SIC data, presumably due to not being in the Edgar, Hoover's or Research Insight database on publicly-traded companies.  Since we are using a rollup SIC, it is easy to provide the general industry without a lot of additional research.  This was completed as of 1/5/06.

Oridinal, Nominal and Numeric Data
----------------------------------
Categorical techniques must be used for SIC analysis since SIC are not ordered at all.

The Yes/No coding is to the best of my belef 1/2, at least from notes made last year when trying to determine Kristen Bachmann's coding.