How many days of pedometer monitoring? Consecutive versus random days of data collection

Abstract

PURPOSE: To determine the number of days and type of data selection (consecutive vs. random) necessary to assess reliable estimates of pedometer data based on 365 days of monitoring. METHODS: Twenty-three participants (7 men, 16 women; age = 38 ± 9.9 years; BMI = 27.7 ± 6.2 kg/m2) wore a Yamax SW 200 pedometer and recorded their step counts for 365 consecutive days. Prior to data analysis, 2% missing values were identified and replaced by the average of remaining values of the person. Nine measurement periods of various lengths were selected. These were 2, 3, 4, 5, 6, 7, 14, 21 and 30 days. Each measurement period was randomly selected 10 times each for consecutive vs. random days from the larger data base. Data were analyzed separately by consecutive and random days. For consecutive days, the starting day was randomly selected from the 365 days of data for each measurement period, and for random days, all the values were randomly selected using SPSS subcommand "select cases". The two-way intraclass correlation coefficients (ICCs) were calculated for the 10 samplings of each measurement period for both consecutive and random days. The mean and standard deviation of those 10 samplings in both cases were inspected to determine the minimal number of days necessary to obtain a reliability of ICC greater than or equal to .80. RESULTS: As the length of the measurement period increased, there was an increase in the average ICC, and a decrease in standard deviation. The average ICCs were larger within consecutive days, compared to random days. The average ICCs for consecutive days ranged from .71 for 2 days to .96 for 30 days, with an average ICCs greater than .80 at 5 days (ICC = .83 ± .04). The average ICCs for random days ranged from .65 for 2 days to .96 for 30 days, with an average ICCs greater than .80 at 6 days ( ICC = .82 ± .04). CONCLUSIONS: The ICC analysis confirmed that at least 5 consecutive days or 6 random days of data collection are necessary to achieve a minimal reliability estimate of .80 in adults' pedometer data.