After the list of variables and their categories were established, the actual coding process occurred. First, each opinion was read and coded according to the variables that were established. Each opinion was read, either in hard copy or on the computer screen, from the Microsoft Word file that had been originally downloaded from Lexis. No analytical software program was used to read or code the cases.
As might be expected, some of the variables
required more detailed reading of the text than other variables did. The year the opinion was filed was typically listed plainly in the header of the respective opinion, so the coding of that variable often did not require any reading of the opinion text itself.
At the other end of the spectrum, some of the other variables required a much more thorough and careful reading of the text. For instance, the variable that reflected what specific activity the provider had been accused of consisted of eight different possible
In many cases, the opinion’s description of the provider’s suspect billing activities alone could consist of multiple pages of text in order to portray the exact occurrences in sufficient detail. In order to code this one variable, these portions of the text had to be scrutinized to ensure that the correct category was chosen to most accurately reflect exactly what the provider in question was accused of. A firm understanding of Medicare claims processing proved beneficial for coding that single variable.
The four variables that reflected the four principal-agent themes of interest also required a particularly careful scrutiny of the text, in order to identify the specific comments and thoughts that needed to be identified. After each opinion was coded, the data on each case was generated from the opinion level data. Using this two step process provided the appropriate level of data to answer each research subquestion.
For instance, Court statements on agency theory were needed at the opinion level, because they reflect the individual thought of an individual Court. Because complete quotations were assessed for this aspect, it would not have been logical to attempt to condense the opinions of multiple courts who wrote an opinion on different motions pertaining to the same case; in fact, it would not have been possible. In contrast, however, reporting opinion level data about the specific activities that providers were accused of conducting would skew those result.
A preliminary assessment of the data was conducted. This preliminary screening served to inform what types of information in the written opinions could be coded. It also served to inform exactly how the list of variables should be structured. The preliminary assessment proved invaluable in determining how the list of variables might be effectively structured. The process used during this preliminary assessment was quite basic and was rooted in common sense.
A great number of the opinions were scanned to collect the basic information in the opinion at hand—such as the specific accusation that the provider faced, the type of provider, the type of service, and any clear principal-agent themes that
might be evident. From out of this process, drafts of the categories for each of the variables were created. This familiarization process was intended to facilitate the actual coding process, after the list of variables had been established. Again, the method
applied in this preliminary data assessment was quite basic. It was similar to the process that any serious researcher undertakes in order to acquaint themselves with a body of unfamiliar data.
Many of the coding choices for each variable are self-explanatory—such as the year the opinion was issued, or the name of the Court that issued the opinion. For other variables, the coding choices were developed after the preliminary data assessment was completed, and this was sometimes supplemented by a working knowledge of the U. S. health care system. For instance, the coding choices for the defendant variable reflect the types of providers who often appeared in the written opinions, and they also represent commonly accepted categories of providers that exist in the U. S. health care system.
The 715 opinions were then screened for their applicability to the research question at hand. Based on established selection criteria outlined in Appendix A, opinions that did not meet the selection criteria were removed from the data set. For instance, some opinions pertained to accused FCA violations solely within the Medicaid program, but because the term “Centers for Medicare & Medicaid Services” appeared within the text of those opinions, they met the Lexis search criteria and were thus included in the initial 715 opinions.
Likewise, some opinions that pertained to defense contracting or other industries also met the search criteria because they briefly cited cases pertaining to Medicare providers who had been accused of FCA violations. Because these opinions, however, did not meet the selection criteria, they were removed from the data set. After all the opinions that did not meet the selection criteria were removed, the entire remaining population of opinions was included in the study; no sampling was used.
The end result was a total of 363 opinions and 233 cases. Due to the manageable size of this dataset, there was no need to use any sampling techniques for this study. Further, because only eight cases were filed between 1966-1986, relying on sampling techniques for this study would not have been advisable. The pre-amendment cohort of eight cases was already too small to describe with percentages. Removing more cases from the pre-amendment cohort by employing a sample would have served only to rob the cohort of meaningful detail.