Cancer Screening Belief Scale. Data Analysis
Before data were analyzed using SPSS 14.0 software, all items were examined to ensure reflection of positive expressions in their corresponding scales. Listwise deletion was used to exclude missing data. Descriptive statistics, item-total correlation, and Cronbach’s alpha coefficients were calculated for each construct.
Structural equation models were used to test separately whether each of the three factors remained stable over time. Items with low loading were dropped. Confirmatory factor analysis was then applied to examine the proposed three-factor model. The purpose of this process was to confirm if there was sufficient empirical evidence suggesting that the model, as specified, might be a viable representation of the true relationships between observed and latent variables (Mueller, 1996). Judgments about model fit were made jointly by assessing the ratio of chi-square to degrees of freedom (χ2/df), root mean square error of approximate (RMSEA), non-normed fit index (NNFI) or Tucker-Lewis Index (TLI), incremental fit index (IFI), and comparative fit index (CFI). The criteria used to determine if the model fit the data were the χ2/df less than three (Bollen, 1989), RMSEA no more than .08 (Raykov, 2001), and values of NNFI, IFI, and CFI at least .90 (Byrne, 1998). Factor loadings were considered statistically significant if the ratio of the factor loading to its standard error was greater than 1.96 or less than -1.96 (Joreskog & Sorbom, 1996). Finally, reliabilities of each of the factors in the cancer screening belief scale were calculated.
Results
A total of 450 survey were distributed to employee and their family members aged 40 and over among the ten participating worksites in Taiwan. Among these people, 375 participants completed and returned their consent form and initial surveys (83%). At the one-month follow up, 304 participants were reached and completed a follow-up survey (81%). Among these, 272 participants completed all items in the cancer screening belief section of the survey and thus were included in the following reliability and validity analyses.
Demographics
The mean age of the participants was 48.18 (SD=8.79), most of them were married (93.3%). About 58.8% were male. Over half of the participants had a college education level or higher (54.0%), and 78.6% had a full time job. Most people (88.9%) indicated their general health condition as “fair” (39.0%) or “good” (49.9%). Even though near half (49.9%) of the participants indicated they had someone in their family who had been diagnosed with cancer of any types, over 90% of the participants rated their perceived risk of getting cancers in the next five years as either the same (49%), low (22%), or very low (20.4%) compared to others in their ages.
Initial Reliabilities
The initial reliabilities for the three constructs measured in the initial 19-item pool showed satisfactory internal consistencies, with Cronbach alphas all greater than .70. The corrected item-total correlations of all the items were greater than .20, and ranged from .32 to .80, indicating that all the pros, cons, and perceived risk items showed sufficient correlations with other items in their corresponding constructs (see Table 1). The correlation matrix among items is available upon request.