(no subject)

[Brittawniolson]

This is a rather broad (and long)  question that I am interested in getting input on from anyone who may have some insights into sampling design and its influence on statistical scale validation. I have a project that is attempting to conduct psychometric validation of a modified community readiness survey.  The sampling strategy (already completed) was driven by budget, input from prevention professionals in the target area and geographic information, and specific desire from the client.  (It is described in greater detail below)   I would like to know what kind of impact this sampling strategy may have on the analysis and what, if any, type of statistical cautions, controls or techniques I should employ to address this influence. The primary objective of the project is to simply validate the scale validity of the survey. The two key objectives: (1) evaluate the internal validity of the domain scales using a the target population and (2) examine the cross-validation of domain scales based on the inter-domain correlations using based on mean domain scores to determine if the average scale scores demonstrated the expected cross-scale relationships.    SAMPLE STRATEGY:   (NOTE - the target is a single county)      Community samples were drawn from six urban neighborhoods, five rural communities and two cultural centers in This County. Once participating communities were identified (by client and prevention professionals), a random sample of directory-listed household addresses was purchased from Info USA. A survey packet was compiled for mailing and included a cover letter, the survey, an incentive drawing card for a cash prize, and a self-addressed, stamped envelope to be used to return the survey. The survey packet was mailed to 5,096 adults in County by The Client.  A reminder postcard was sent to the entire sample 10 days after the initial mailing. To increase response rate respondents were given the opportunity to enter a drawing for $150.00 with the return of a completed survey. (Full anonymity of survey responses was guaranteed)   Urban Sample Selection Due to the fact that urban neighborhoods do not conform to census tracts or grid lines, a broad conception of neighborhood was used to select the six urban groups. The urban target area has six public high schools. The schools served as geographic center points for sample selection. For each high school area, a random sample of approximately 500 individuals were selected from resident households within a 1.5 mile radius of each school (n=2,999). Prevention specialists indicated that the various areas of the city represented by the high school groupings would in fact represent different levels of readiness - thereby ensuring variability in the primary variable of interest. Finally, in order to ensure adequate representation of the urban site's growing minority population an additional 100 surveys were distributed to the Hispanic and Vietnamese Cultural Centers. This approach was designed to ensured adequate representation of all areas of the city and provided a random sample within each identified geographic cluster while also ensuring variability in the dependent variable (community readiness).   Rural Sample Selection Between 250 and 600 adult residents were randomly chosen from each of five rural communities based on the community's total population (n=1997). The five rural communities were selected based on the input of prevention professionals who worked in the rural communities of the County. The five selected communities represented a variety of levels of readiness based on the perceptions of prevention professionals.     Sample Characteristics              The total number of participants in the survey was 1,058 providing an overall response rate of 21%. Urban neighborhood response rates ranged from 16% to 23%; rural response rates ranged from 18% to 23% ; and the cultural center response rate was 82%.  In total, 62% of the sample were urban respondents, 37% were rural. Additionally, a comparison of selected respondent characteristics with similar county demographics based on the 2000 Census demonstrated that the survey sample reflected the adult population distribution fairly well. For example, 89% of the sample was Caucasian compared to 94.5% of the county population and 29% had a college degree compared to 33% of county population. However, the sample composition included slightly more females (55% versus 50% population) and more seniors age 65 or older (26% versus 10% of population). The overrepresentation of females was not unexpected as this group is frequently overrepresented in general population surveys; however, the overrepresentation of seniors was somewhat surprising because it was counter the typical trend for population surveys (Groves, 1989).  The overrepresentation of senior respondents may reflect their concern about personal safety; there is an abundance of research indicating that senior citizens tend to feel particularly vulnerable to crime (see e.g., Doerner & Lab, 1998). Nonetheless, generalizations to the entire population should be made with extreme caution.   Based on the above sampling design and the goals of the project, what advice, suggestions, cautions would you provide with regard to scale validation?   Thanks in advance for any input you may be able to provide. Please feel free to respond to me off-list by simply replying to this email directly.                                                                                                                  Brittawni Olson Go Confidently in the direction of your dreams. Live the life you have imagined. Henry David Thoreau (1817-1862)