Two context effects did not yield significant influence on the use of probability expressions in this study. The insignificant finding might have been confounded by other factors which were not well controlled by the instrument. The effect of the magnitude of numbers in the fractional restatement could have been confounded by the percentage presented before the restatement. The expectation effect could have been confounded by the different percentages reported in the three statements, which represented the same probability interval. Future research needs to verify these context effects by eliminating or controlling the confounding variables.

The rank order of the 11 probability expressions which has been found in the literature (Reagan et al., 1989) was supported both by the logit scales (except the middle point) and also the numeric-to-verbal mapping within the educational context. AAlmost impossible," Aan even chance," and Aalmost certain" served as good anchors for the lowest, middle, and the highest probabilities respectively, as evident from subjects¹ processing of numeric-to-verbal mappings within a given context. Expressions incorporating the stem "likely" and those incorporating the stem Aprobable" appeared to mean similarly to subjects; this again verified the finding by Reagan et al., (1989). However, "likely" was chosen more often than "probable" by subjects in both samples. "possible" and "likely" were used to similar extent by subjects. Since these two words were found to cover a broad range of probabilities, they may not be a good candidate for conveying specific information about uncertainty, especially in high stake situations such as interpreting achievement test results or diagnosing intelligence of young children.

Rankings of 11 expressions by the logit scale were reliable and could be directly explained. However, the logit scale values make sense only if they are transformed into probabilities to show the relative greater likelihood of one expression over another. As for the least squares scales, doubt was cast over the validity of scale values within a proximity to the origin. This issue awaits future research to clarify.

There was no difference between the 1992 and the 1998 findings regarding the context effects, numeric meanings of probability expressions, and paired comparison scales. This observation suggested consistency in subjects¹ behavior between years.

The subjects of both studies were volunteer undergraduates who enrolled in education courses in a large mid-western university. Since a large proportion of subjects were education majors who gave very similar responses in 1992 and 1998, we felt confident to conclude that our results can be at least generalized to prospective teachers in training at state universities. Future researchers pursuing the same line of research may wish to recruit subjects from diverse backgrounds and age ranges so as to expand the believability and generalizability of our two-time findings.