ASSESSING THE IMPACT OF PHYSICAL CONDITIONING, DIETARY INTAKE, BODY FAT, AND TOBACCO USE ON BLOOD PRESSURE PARAMETERS: A TWO-METHOD MEASUREMENT DESIGN APPROACH
Open Access
- Author:
- Olchowski, Allison Elizabeth
- Graduate Program:
- Biobehavioral Health
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- June 12, 2007
- Committee Members:
- John Walter Graham, Committee Chair/Co-Chair
Robert James Turrisi, Committee Member
Sheila Grace West, Committee Member
Edward A Smith, Committee Member - Keywords:
- two-method measurement design
missing data
cost-effectiveness - Abstract:
- The two-method measurement design was applied to answer substantive questions pertaining to hypertension and several lifestyle-based risk factors; specifically, analyses involved estimating the cross-sectional association of three blood pressure parameters (systolic blood pressure (SBP); diastolic blood pressure (DBP); and pulse pressure (PP)) and four known hypertension risk factors (physical conditioning; dietary intake; body fat; and tobacco use). The two-method measurement design, a recent statistical advancement in the area of planned missingness approaches, measures constructs using several indicators of variable cost and validity. Cheaper, less valid measures of a construct and more expensive, valid measures of the same construct collectively serve as manifest indicators. All participants provide data for the cheap measures; a small proportion of participants also provide data for the expensive measures. When at least a subsample of participants provide complete data, bias correction models allow for the modeling of measurement bias (i.e., reduced construct validity) associated with the cheap measures; resulting parameter estimates are efficient and unbiased. A simulation paradigm was used to apply the two-method measurement design to empirical NHANES data. The performance of the two-method measurement design was compared to that of financially-equivalent complete case designs. For two of the four predictors – body fat and tobacco use – application of the two-method measurement design produced statistical power advantages beyond those yielded by the financially-equivalent complete cases models. Under a hypothetical budget constraint of $20,000, complete case body fat data could be collected from N=333 participants; however, the two-method measurement design behaved as if the complete case sample sizes were N=1469, N=1625, and N=1565 for testing the effects between body fat and SBP, DBP, and PP, respectively. Under the same budget of $20,000, complete case tobacco use data could be collected from N=363 participants; the two-method measurement design behaved as if the sample sizes were N=1513, N=655, and N=872 for testing the effects between tobacco use and SBP, DBP, and PP, respectively. Application of the two-method measurement design was comparatively less effective for the physical conditioning and dietary intake variables. A potential explanatory factor involves the general lack of association between cheap and expensive measure indicators for these two variables. In general, the strength of association between the independent and dependent variables was inversely correlated with the increase in statistical power produced by the two-method measurement design (results consistent with previous research). Results have implications for future application guidelines. To maximize the utility of the two-method measurement design, cheap and expensive measures for a given construct should be highly correlated. It is recommended that researchers collect small amounts of data from candidate cheap measures to determine, a priori, the set of cheap measures that best correlates with expensive measure data. It is also helpful if researchers are able to anticipate, to some degree, effect sizes between independent and dependent variables of interest; this offers researchers the opportunity to more accurately tailor data collection to achieve maximal cost-effectiveness. Recent interest in the efficiency of health prevention programs, as well as limited external funding sources, has placed an increased emphasis on cost-effective research within many behavioral health disciplines. Whenever researchers have the opportunity to collect data for a particular construct using several measures of variable cost and construct validity, the two-method measurement design offers the potential for cost-effective data collection and unbiased and efficient parameter estimation.