Methodology

Methodology

The survey data presented in CHRT’s Physician Survey briefs were produced from a mail survey of primary care physicians practicing in Michigan. The physician samples were randomly generated from the American Medical Association (AMA) Physician Masterfile, a comprehensive list that includes both AMA members and non-members. Physicians who responded but reported they were no longer practicing primary care were removed from the analysis.

Given the sampling methods used in the physician surveys, some respondents may have been included in multiple survey years; because of the de-identified nature of responses, it was not possible to identify physicians who responded to more than one survey.

  • Physician Survey 2018: The survey was conducted between August and December 2018 and included a sample of 1,800 primary care physicians in Michigan. The sample was composed of a random statewide sample of 1,500 physicians and a rural oversample of 300 physicians. Rural was defined as Micropolitan and Noncore Counties, according to the 2013 NCHS Urban-Rural Classification Scheme for Counties. Potential respondents received up to three mailings, with $10 included in the first mailing to encourage response. The mailing included information about how to complete the survey online in Qualtrics, rather than by hard copy, if respondents preferred this option. Both surveys returned by mail and online were merged to create a final data file. The final sample included physicians from two primary care specialties: family medicine and internal medicine. The survey had a response rate of 42 percent and a margin of error of ±4.0 percent. There were 588 physicians in the final analytic sample. To account for the oversample, results for the 2018 survey are weighted on urban/rural classification and specialty.
  • Physician Survey 2016: The survey was conducted between July 2016 and October 2016 and included a sample of 1,500 primary care physicians practicing in Michigan. Potential respondents received up to three mailings, with $10 included in the first mailing to encourage response. The mailing included information on how to complete the survey online via Qualtrics, rather than by hard copy, if respondents preferred this option. Both surveys returned by mail and online were merged to create a final data file. The final sample included physicians from two primary care specialties: family medicine and internal medicine. The survey had a response rate of 47 percent and a margin of error of ±2.5 percent. There were 603 physicians in the final analytic sample.
  • Physician Survey 2014: The survey was conducted between December 2013 and April 2014 and included a sample of 1,000 primary care physicians practicing in Michigan. Potential respondents received up to three mailings, with $5 included in the first mailing to encourage response. The final sample included physicians from two primary care specialties: family medicine and internal medicine. The survey had a response rate of 36 percent and a margin of error of ±5.5 percent. There were 317 physicians in the final analytic sample. Physicians who reported that they were unsure whether they participated in an innovative compensation model or that they were not participating at the time of the survey but planned to do so in the future were considered as non-participants for the purpose of this analysis.
  • Physician Survey 2012: The survey was conducted between October 2012 and December 2012 and included a sample of 1,500 primary care physicians practicing in Michigan. Potential respondents received up to two mailings, with $5 included in the first mailing to encourage response. The final sample included 500 physicians each from three primary care specialties: pediatrics, family medicine and internal medicine. The survey had an overall response rate of 54 percent and a margin of error of ±3 percent. There were 714 physicians in the final analytic sample. Final results were weighted to adjust for non-response in each of the three primary care specialty groups.

Results were analyzed using the most recent versions of SAS and SPSS software commercially available at the time of the publication of the brief. Statistical significance of bivariate relationships was tested using z tests, odds ratio tests, or chi-square tests. Unless otherwise noted, all results reported were statistically significant at the p < 0.05 level.