Bachman, L. F. (1982). The trait structure of cloze test scores. TESOL Quarterly, 16(1), 61–70.
Population: Two groups of non-native English-speaking students entering the University of Illinois. Age range was 17 – 47 with a median age of 25.
Size: Fall term, 1980 (N = 360). Spring term 1981 (N = 102)
Details: The cloze test was given as part of a placement test battery at the beginning of the semester. This battery also included a 100 item multiple-choice test of structure and a 130-word dictation test. Students were given 20 minutes to complete the cloze test. In this study, the “trait structure of a cloze test was examined using confirmatory factor analysis. A cloze passage with rationally selected deletions [RD’s] of syntactic and cohesive items was constructed.” The structure had three specific and one general trait, deemed “best to describe the data.” In specific, a 365-word passage from an introductory level textbook in social psychology was selected for its appropriateness of difficulty level and "represetativeness" of content. In order to test hypotheses regarding the level of language context measured by the cloze test, three types of deletions were made: (1) syntactic, which depends only on clause-level context, (2) cohesive, which depends upon the interclausal or intersentential cohesive context, and (3) strategic, which depends on parallel patterns of coherence. These deletions provided 11 syntactic items, 15 cohesive and 4 strategic, for a total of 30 items with an average deletion ratio of 1:12. An acceptable alternative scoring procedure was used, with the key based on the responses of a native speaker pre-test group and on the acceptable alternatives provided by Hassan and Halliday’s (1976) description of cohesion.
Conclusions: The results suggested that a modified cloze passage using RD [rational deletions] is “capable of measuring both syntactic and discourse-level relationships in a text, and that this advantage may outweigh considerations of reduced redundancy which underlie random deletion procedures.” However, since native speakers tested with such great variance in their ability to negotiate language at the lower levels (see also Alderson, 1980), the conclusion was reached that the cloze test may be better suited for higher-level testing.
Comments: My current conclusion is that the C-test format of the cloze test is the best to date for assessing NNS proficiency levels, especially when NS performance data are available on the same texts.
 Confirmatory factor analysis tests “causal structure” among a number of variables (Jöreskog, 1969 & 1978).
 Klein-Braley (1982) modified the cloze to a paragraph-style testing format where the last sentence(s) are kept in tact and the beginning truncated at the rate of every second word starting with the second word of the second sentence. (Truncated meaning the last half deleted and replaced with underscores at the rate of 1 underscore for each character deleted).