Inventory Development
 Test-Retest Reliability


 The alpha reliability procedure does not address an important source of error-that due to fluctuations in individual responses across time. This last concern addresses the "stability" of both the construct of stress and the measures used to assess it across time (Gable, 1986). Limited test-retest reliability data have been collected for the initial form of the TSI and are reported in Fimian (1986c).

To establish the test-retest reliability, two sets of TSIs were mailed to a random sample of 60 North Carolina special education teachers in February 1985 (i.e., initial sample) and again within 2 weeks (Samples 1 to 4). This sample was generated from computerized personnel lists acquired from the North Carolina State Department of Public Instruction. Of those completed and returned, 82% or 49 from both the initial and the retest samples proved usable and were included in the analyses. When the surveys were initially distributed, four different sets of administration directions were included, one each to 25% of the sample. These directions requested that the teacher complete the second TSI after 2 hours (Sample 1, n = 9), 1 day (Sample 2, n = 15), 1 week (Sample 3, n = 11), or 2 weeks (Sample 4, n = 14) after completing the first. Of the completed inventories, 82% or 49 from Samples 1 through 4 were included in the analyses. Paired samples t-tests were used to determine the correlations between the initial and four retest samples (Nie et al., 1975). Test-retest estimates for the subscales and scale are reported in Table 17 . Correlations for the 2-hour to 2-week samples ranged .42 to .99 for subscales and.67 to.99 for the whole scale. Only 2.5% of the subscale correlations proved insignificant; the majority of the remaining test-retest reliabilities correlated beyond the .001 level, demonstrating 45% to 98% shared variance across 2-hour to 2-week intervals. Thus, the test-retest reliabilities proved acceptable for future research and time series research.

Independent evidence of the test-retest stability of the TSI is provided from data collected from a sample of 39 Georgia teachers, first in September of 1984 and then again in November of 1984. These data, presented in Table 18 , indicate that test-retest correlations ranged.49 to.84 (p = .001) for the TSI subscales and .76 (p = .001) for the Total Stress Score. Thus, all TSI subscales, and the TSI scale for the strength dimension, were significantly related to their analogous measures across 8-week intervals (Fimian, 1986c).