Findings from the QMA Survey

Rebecca L. Schewe
Assistant Professor of Sociology
Research Fellow, Social Science Research Center
Mississippi State University

A mail survey focusing on mastitis management and antimicrobial use was sent to 1,700 dairy producers in Michigan, Pennsylvania, and Florida between January and February 2013. The final survey totaled twenty pages, including the full-color front and back covers, with seventy-one questions and an additional lined page for written comments. The survey was pretested by co-PIs and Extension agents from all three states and then face-to-face pre-tests were conducted with fifteen dairy producers in Michigan and Wisconsin. Pre-tests indicated that average time to completion was less than fifteen minutes. Survey administration followed strategies recommended by Dillman and colleagues (2008, Babbie 2009) to maximize response rate including: first class postage, full-color covers, personalized addresses, a distinctively sized and envelope with color picture, five points of contact, and a two dollar bill.

The first mailing was sent on January 17, 2013 and included a cover letter, survey, and $2 incentive. The second mailing was sent on February 1, 2013 to all non-respondents and included a full-color personalized reminder postcard. The third mailing was sent on February 8, 2013 to all non-respondents and included a second copy of the cover letter and survey. The fourth mailing was sent on February 15, 2013 to all non-respondents and included a second copy of the reminder postcard. The final fifth mailing was sent on February 25, 2013 to all non-respondents and included a third copy of the cover letter and survey.

cow-color-qma-type-webThe survey was sent to a stratified random sample of USDA Grade A certified dairy farms in Michigan, Pennsylvania, and Florida. These three states were selected to represent the maximum variation in dairy farm size and management, with Florida having the largest average farm size of 284 cows and 88% of farms having more than 500 cows (USDA Ag Census, 2007: Table 11). Pennsylvania has the smallest average farm size of 66 cows and only 10% of farms having more than 500 cows (ibid), with Michigan falling in the middle with an average of 130 cows and 43% of farms having more than 500 cows (ibid). Due to the small number of dairy farms in Florida, all 128 Grade A herds were included in the sample. The sampling frame for Florida was obtained through co-PI Dr. Ray Mobley and included all currently certified Grade A herds, names and addresses.


In Michigan, the sampling frame was obtained through Freedom of Information Act (FOIA) requests to both the Michigan Department of Agriculture and Rural Development (MDA) and the Michigan Department of Environmental Quality (MDEQ). The MDA provided the names and addresses of all current Grade A certified dairy farms. Although we requested herd size to be included, the MDA declined this portion of our FOIA request citing privacy concerns. This was a concern because without herd size, stratification would not be possible. Fortunately, the MDEQ provides the current and pending Concentrated Animal Feeding Operation permits available as a spreadsheet on their website. By comparing this spreadsheet which includes name, address, type of animal operation (i.e. dairy), and animal units, we could identify those farms with more than 700 cows (the MDEQ requirement for a permit). Once the 98 farms over 700 cows were identified, I sent the updated sampling frame to Michigan State University’s Dairy Team (including vets, faculty, and extension agents). I asked team members to highlight any farms that they knew to be over 500 cows. This added an additional 8 farms to the large-farm category. Once I began to draw the sample, one duplicate farm was discovered. This farm was duplicated on the small and large lists, so it was removed from the small farm sample. The resulting sample included all 106 large farms and 640 small farms.


In Pennsylvania, the sampling frame was obtained through co-PI Dr. Ernest Hovingh and an FOIA request to the Pennsylvania Department of Agriculture (PDA). Dr. Hovingh, an Extension veterinarian, had a confidential list of Grade A certified farms including herd size. Hovingh assigned unique identifiers to each farm and provided an anonymized spreadsheet that included only identifiers and herd size. I used this anonymized spreadsheet to define the strata and draw the sample. Based on the distribution of herds, I selected 250 cows as the cutoff for the “large farm” sample. This is significantly smaller than the cutoff for MI herds because PA herds are left-skewed. The FOIA request provided a spreadsheet of all CAFO permits from the PDA including farm name, address, type of animal operation, and animal units. Hovingh proofed the sampling frame by comparing the reported number of lactating cows to the CAFO list and to his and other extension agents’ industry knowledge, as well as to publicly available data online. He corrected several erroneous herd size numbers and filled in several missing data points for herd size (adding 5 farms to the sampling frame and moving 39 farms from the “general population” into the “large farm” category based on corrected herd size). The resulting sampling frame included 7133 farms. 1010 were removed from the sampling frame before a sample was drawn, 767 of which had no data reported for herd size and 243 of which had a reported herd size of zero cows.

Both Michigan and Pennsylvania samples were stratified by farm size due to the small number of large farms in those states. Previous studies have linked farm size to Bulk Tank Somatic Cell Count (BTSCC) and mastitis (Norman, Cooper, and Ross 2011), so a stratified sample was important to guarantee adequate representation of large farms in the final sample. Strata were defined based on the farm size distributions of each state, in Pennsylvania large farms were defined as those over 250 cows and in Michigan large farms were defined as those over 500 cows. All farms in the large farm strata were included in sample (see Table 1 for sampling weights and strata details). Because the strata varied in both size and sampling ratio, it is necessary to appropriately weight the resulting data to ensure that the data represent the population and do not overrepresent the over-sampled strata (in this case the large Michigan and Pennsylvania farms and all Florida farms). Sample weights for the five strata were defined as probability weights (pweight = 1/sampling ratio).

2013 05 14 – Report on Mastitis Management and Labor Management

Accompanying Tables

Leave a Reply

Your email address will not be published. Required fields are marked *