Inventory Development
 Early Factorial Work


Following procedures suggested by Nunnally (1978) and outlined by Gable (1986), a number of studies were conducted that would examine the factorial validity of the Teacher Stress Inventory. The first, conducted on the Connecticut data collected using the Teacher Stress Scale pilot form, resulted in a six-factor solution (Fimian, 1985). Then, with the additional items added to the pilot form, the second set of factor analyses was conducted using this pilot form to collect data from the Vermont teachers; this, too, resulted in the six-factor version of the TSI form (Fimian, 1984b).


These teachers, representing 21 individual samples from eight states, provided data for the most recent round of TSI development activities. Sample designations, descriptions, and related information are presented in Table 10 . Of these, 13 samples' inventories were distributed through the mail; the balance of the data were collected either at workshops or through regional surveys. More detailed information regarding these samples can be found in Fimian (1983), Courtney (1987), Honaker (1987), and Zacherman (1984). Of these 3,401 teachers, 743 were included in all regular teachers" samples, 1,778 were included in "all special education teacher" samples, and 880 were grouped in combined special/regular teacher samples; of this last group, 88 were unclassifiable, and the majority were special education teachers. Thus, the final counts were 960 (regular education), 2,353 (special education), and 88 (unclassified by group), for a total of 3,401 teachers.

The majority of the teachers in the norm sample were female (n = 2,56 1; 75%), and the balance were male (n = 726; 21 %; some percentages may not sum to 100% due to missing cases and/or rounding error). Many were in their twenties (n = 1,292; 38%), with the remainder in their thirties (n = 1,398; 41%), their forties (n = 386; 11%), or fifty or older (n = 164; 8%). They included secondary school teachers (n = 1,420; 42%), elementary school teachers (n = 791; 23%), or middle school teachers (n = 499; 15%). The majority had less than 10 years' experience (n = 2,092; 62%), whereas the balance reported more. The minority reported teaching fewer than 20 students per day (n = 1,008; 37%), and the rest reported more than 20 (n = 1,728; 51%). A minority had achieved a bachelor's degree (n = 450; 13%), the balance an advanced degree.

Statistical Analyses

Based on the teacher data, and in order to identify stress factors, preliminary principal components factor analyses were conducted and followed by oblique and varimax rotations using the stress strength data collected in questions 1 through 49, according to an instrument development model proposed elsewhere (Child, 1970; Nie, Hull, Jenkins, Steinbrenner, & Bent, 1975) and later expanded upon by Gable (1986). Then, the internal consistency reliability estimates for the TSI subscales and scale were examined using Cronbach's coefficient alpha (Hull & Nie, 1981; Nunnally, 1978). Once valid and reliable TSI subscales and scores were identified, the relationships among these were investigated using Pearson product-moment correlational analyses. The factorial validity of the revised TSI was examined using the 49 TSI items. Preliminary principal components analyses were conducted and followed by oblique and varimax rotations. Based on the 49-by-49 item intercorrelation matrix and the principal components analyses, 10 factors for the stress strength dimension emerged that accounted for 58% of the stress variance; only factors with eigen values exceeding 1.0 were retained. An initial inspection of the factor patterns indicated 10 discrete factors.

Then, one set of selection criteria was applied to each of the 49 items. Items were retained that (a) had factor loadings of .35 or greater on the stress strength dimension, (b) loaded clearly on only one factor (i.e., simple structure was achieved), and (c) contributed to the subjective interpretability of the particular factor on which the item loaded. Once the factors were identified, the TSI's internal consistency or alpha reliability estimates were then examined. One estimate was generated for each of the 10 stress strength subscales. Also, one estimate was generated for the total group of items. Then, any item that did not reduce the internal consistency reliability of the particular subscale in which it was nested was retained. Factors whose alpha reliability estimates exceeded .60 were retained; exploratory alpha ranges of .60 to .90 for the TSI subscales and .85 to .95 for the TS1 scale were targeted. Final acceptance or deletion of the TSI items, therefore, was based on a combination of findings from each of the analyses. Items were kept that were not only valid in terms of subscale/scale factorial validity, but also reliable in terms of subscale/scale internal consistency reliability.


Table 11 contains the 49 retained and abbreviated item stems with their communalities and component loadings derived from the oblique rotations for the stress strength measures for each of 10 resulting factors. Employing a root criterion of unity, the 10-component solution derived from the strength item scores accounted for 58% of the total stress strength variance associated with the item interrelationships. It was evident from Table 11 that (a) 10 discrete and interpretable factors resulted and (b) of the original 49 items, all exceeded the .35 loading criterion with all but 2 exceeding .40. Thus, no items were deleted from the pool. All items included in the initial form of the TSI were retained for the final form.

The percentage of explained variance per factor was then calculated by summing each of the squares of their correlation coefficients listed on the varimax factor correlation matrix and then dividing this sum by the number of entries (i.e., 49); per-factor contributions were then summed. Lotus 1-2-3 (Version 2.0) was used to conduct all calculations.

Since only the stress strength dimension was under investigation, one total score for that dimension was developed using the item mean data. Also, since the stress strength dimension is collectively defined in terms of the 10 factors, item mean scores were used to develop 10 conceptually similar subscales. Subscale scores were derived by first summing the item scores for the stems nested within each subscale and then dividing the resulting value by the number of items in that particular subscale. In this fashion, each subscale's score falls within the 1-to-5-point strength range. Thus, the relative strength of each collective body of stressful events (hereafter termed "subscale') can be easily interpreted. Insofar as the overall stress experienced by teachers is operationally defined as the relative strength with which all 49 events are experienced, the 10 stress strength subscale scores were first summed and then divided by the total number of stress factors for a Total Stress Strength Score. By so doing, the resulting Total Stress Strength Score should fall within the 1-to-5-point strength range, and the relative strength of stressful events can be easily interpreted.