One of the primary measures of
reliability is that known as alpha or internal consistency reliability.
Gable (1986) noted that this reliability coefficient is drawn directly
from the domain-sampling theory of measurement error described in
Nunnally (1978). In so doing, the use of Cronbach's (1951) coefficient
alpha addresses the error due to item sampling practices and the use of
"one shot," single, or cross-sectional administrations of an
instrument. Also, it provides an index that allows users to estimate the
degree to which the items within a subscale or scale "hang
together." Poorly selected items generally do not relate to one
another as strongly as do items that have undergone a thorough and
systematic process of item selection. As measurement error "creeps
in," the body of items relate to one another less and less;
accordingly, the alpha reliability estimate that represents these
relationships tends to be lower than is conventionally desirable. In
short, the less "internal consistency" within a body of items,
the lower the alpha reliability estimate. The more inconsistent the
response patterns are across items within an item pool, the lower the
Evidence of the TSI's internal consistency is provided in Table 16 . This table contains the names, items, and alpha reliability estimates for each derived TSI subscale and scale, once each for the special (n = 2,352), regular (n = 962), and combined (n = 3,401) teacher samples. Subscale reliability estimates for the special education teachers ranged from a low of .67 (Professional Investment) to a high of.86 (Gastronomic Manifestations); one estimate fell in the .60s, three in the .70s, and six in the .80s. The whole scale alpha reliability was .93. Subscale reliability estimates for the regular education teachers ranged from lows of .70 (both Professional Investment and Fatigue Manifestations) to a high of .87 (Time Management); no estimates fell in the .60s, six in the .70s, and four in the.80s. The whole scale alpha reliability was.92. Subscale reliability estimates for the combined teacher sample ranged from a low of .75 (Professional Investment) to a high of .88 (Gastronomic Manifestations); no estimates fell in the .60s, two in the .70s, and eight in the .80s. The whole scale alpha reliability for the combined sample was .93.
Since all items were included in the reliability analyses, and none reduced the subscale or scale reliability estimates, all 49 items were again retained. Based on these data, also, the pre-established target range of .60 to .90 for all TSI subscales was reached in each instance; the smallest approximated .67, whereas the largest exceeded .88. Sixty percent of the estimates fell above .80, and 96% fell above .70; thus, it is evident that the samples responded to the TSI in a consistently reliable fashion. Also, and with respect to the whole scale alpha estimates of .93,.92, and .93 for the combined, special education, and regular education teacher groups, respectively, these values indicate a high degree of overall internal consistency across samples.
Based on the aforementioned criteria for item acceptance or deletion, item numbers were neither reduced via the factorial validation nor reduced via the internal consistency reliability analyses. Thus, all 49 stress strength items were retained, and all reliabilities were judged adequate for present subscale inclusion in the TSI.