Scheduled for Poster Session: The Study and Assessment of Activity and Performance Across Disciplines, Friday, April 11, 2008, 8:45 AM - 10:15 AM, Convention Center: Exhibit Hall, Reseach Consortium Poster Sessions


Cross-Validation of Individual Information-Centered Methods for Handling Step Count Missing Data

Minsoo Kang1, David A. Rowe2, Tiago V. Barreira1, Terrance S. Robinson3, Charles D. Kemble3 and Matthew T. Mahar3, (1)Middle Tennessee State University, Murfreesboro, TN, (2)University of Strathclyde, Glasgow, United Kingdom, (3)East Carolina University, Greenville, NC

The purpose of this study was to cross-validate individual information-centered methods for handling missing data using adults' and older adults' step count data sets. 7-day Yamax SW-200 data from 118 adults and 91 older adults were screened, and only the ones that had at least 1 weekend day and at least 3 weekdays were kept. A semi-simulation approach, proposed by Kang et al. (2005), was used to create data sets, taking the data characteristics into consideration in determining the effective recovery method. The data from adults and older adults were analyzed separately. The missing values were replaced with the following two individual information (II)-centered and two group information (GI)-centered methods: a) Average of remaining days of the participant; b) Average of remaining weekdays or weekends depending on the type of missing day by the participant; c) Average of the same day (e.g., Monday) of all participants; and d) Average of the type of missing day (weekday, weekend day) of all participants. The impact of various combinations in recovering missing values was evaluated by comparing the known values purposely removed from the data set with the replacement values based on the four recovery methods. Root mean square difference (RMSD), in which the differences between the original and replacement values were squared, averaged and square rooted, was used to determine the effectiveness of the various recovery methods. A smaller RMSD represents a more accurate recovery of the missing values. Paired t-tests were also used to examine the mean differences between the original and the replacement values. Overall, the II-centered methods showed smaller RMSDs than the GI-centered methods. More specifically, the replacement condition “a” provided the minimum RMSD: 4,444 for adults (b = 4,549; c = 5,297; d = 5,255) and 1,620 for older adults (b = 1,898; c = 3,790; d = 3,610). There was no significant difference between the original and replaced values for condition “a” for adults, t = 1.17, p = .24 and older adults, t = -1.26, p = .21. In conclusion, II-centered methods produce better results than GI-centered methods for handling step count missing data, and substitution of missing data points by the average of remaining days appears to be the most accurate missing recovery method for adults' and older adults' step count data.
Keyword(s): exercise/fitness/physical activity, measurement/evaluation, research

Back to the 2008 AAHPERD National Convention and Exposition (April 8 - 12, 2008)