A: The majority of research dealing with issues of family violence has relied either on clinical/legal samples such as women and children in shelters and people required to attend violence intervention programs or convenience community samples that may be recruited through newspaper advertisements. Although we've learned a lot from research based on those samples, there are limitations to that kind of sampling that make it impossible to generalize findings to a more representative part of the community. We wanted to see if RDD could be used to recruit more representative samples for our laboratory studies.

A: To create the phone numbers we matched a central office code (the first three digits in a seven-digit phone number) with a randomly generated four-digit line. We began by downloading a list of all central office codes from the North American Numbering Plan Administration website and locating the central office codes used in our county that were no more than a 45-minute drive from our university lab and were not assigned to mobile phone providers. That process left us with 216 central office codes. We then created a grid in which each row contained a randomly sorted central office code and five columns with a randomly generated four-digit number in each column. We generated approximately 230,000 rows and printed the grids for our telephone interviewers.

For each row, interviewers used as many of the five four-digit numbers as needed to conduct an interview. If the first call on that row resulted in a completed interview, then the caller moved to the next row for the next call. If the first call was a non-working number, a business number, a refusal/hang-up, or a number where no one answered after four calls, the caller stayed on the same row but moved to the next column. If the number dialed was a busy signal or a ringing but unanswered phone, the caller temporarily skipped that row and waited at least 15 minutes to call the number again. Those numbers were called a maximum of four times. Interviewers also moved to the next row if they used all five columns without completing an interview. Every call was written in a calling log and was later entered into a computerized database. Now, of course, dialing and data recording is all computerized and automated.

A: Our interviewers were undergraduates who earned research credits by working in our lab. They worked in 1.5-hour blocks, and their first six sessions were devoted to training. They had an orientation to RDD and the methods for data recording and collection during the first two sessions. We introduced the students to the telephone script and clinical surveying in the third and fourth sessions and then they spent the fifth and sixth sessions practicing the script through role playing. None of the students were allowed to make actual phone calls until they went through an entire role play smoothly.

During each calling session, interviewers were supervised by our staff members with a ratio of 1:6. We conducted about 14 calling sessions per week. Between September 1998 and May 2002, our undergraduate research assistants logged 12,053 hours.

A: We compared the demographics of all phone respondents regardless of their eligibility for our laboratory study to the 2000 U.S. census data for our county. We found small differences between our respondents and the general population, in that we had lower numbers of respondents who identified as Latino/Hispanic, "other race," or two or more races. African Americans and American Indians were slightly overrepresented in our sample as were married couple households and men and women 30 to 44 years of age. We also slightly oversampled middle income families, but overall our sample was fairly representative of the county population. Fewer people identifying as Latino/Hispanic in our study could be attributed to our study being conducted only in English, and oversampling African American and American Indians could be seen as a benefit because our county's proportions of those groups are less than the national proportions. We felt that none of these small differences affected the representativeness of our targeted samples of couples parenting a young child and of couples married or living together.

A: When we compared the two groups in our parenting study, they did not differ substantially, but they did show some differences in income, couple conflicts, and corporal punishment. Laboratory participants in the family study were less likely to earn over $100,000 a year and reported using a wider variety of corporal punishment as well as more frequent relationship conflicts. Couples who participated in our couple study differed significantly from qualified respondents on only one demographic factor. Distressed/aggressive couples were slightly more likely to have the man be the RDD-screened respondent; the effect size was small. We believe that these results provide some evidence that researchers wishing to generalize to the larger general population can use RDD to generate representative participant pools for laboratory studies.

A: One advantage was that we were able to spread the recruitment out over several years so that we could have a steady level of participants in the laboratory without long waiting periods for the respondents. An outside firm probably wouldn't have been interested in that kind of timeframe, and even if they were agreeable to it, the cost would have been too much for us. We were able to fund two complete studies for what it would have cost to pay a company for recruitment. Also, doing the RDD ourselves allowed us to open and close study groups easily depending on the needs of the laboratory studies.