Behavior Research Methods, Instruments, & Computers 1997,29 (3), 401-424 Unprimed stem completion is only moderately predicted by word frequency and length RAYMOND J. SHAW West Virginia University, Morgantown, West Virginia Factors that determine unprimed performance in word-stem completion were investigated. Re­ searchers' descriptions of materials used in word-stem completion suggest that normative word fre­ quency, number of alternative completions, and response length are believed to be important. A total of 160 students provided normative responses to each of 914 multiple-completion three-letter word stems, resulting in over 12,000unique responses. Although the number of alternative responses made correlated well with the number possible, word frequency and length were extremely variable in their relation to response frequency across stems. Creating more precise materials for implicit memory stud­ ies appears to require consulting nonnative response data. Recently, much speculation, research, and debate has (e.g., perceptual versus conceptual processing at test; Ja­ focused on issues ofimplicit memory (see, e.g., Graf & coby, 1983; Roediger & Blaxton, 1987), imprecision of Masson, 1993; Lewandowsky, Dunn, & Kirsner, 1989, testing technique (Jacoby, 1991; Toth, Reingold, & Jacoby, for reviews). Implicit memory is revealed when perfor­ 1994), variation in experimental design (e.g., within- ver­ mance on a task is affected by a previous experience, but sus between-subject designs; A. S. Brown & Mitchell, the person need not reflect back to that experience to 1994; Challis & Brodbeck, 1992), and insufficient sam­ complete the task. This is in contrast to explicit memory, ple sizes to detect significant differences (e.g., A. S. which is revealed by traditional tests ofmemory that ex­ Brown & Mitchell, 1994; Chiarello & Hoyer, 1988). It is plicitly refer to the previous experience (e.g., cued re­ interesting to note that this list of factors covers most of call). One ofthe central issues is to understand what under­ Jenkins's (1979) tetrahedron offactors (subject, encoding, lies differences between implicit and explicit memory. retrieval, and materials variables) to consider in the study Two contrasting, dominant viewpoints have been that ei­ ofhuman memory. ther separate memory systems are involved (e.g., Tulving One ofthe tasks that has been used frequently in implicit & Schacter, 1990), or different types ofprocessing are re­ memory research is word-stem completion. In that task, quired (e.g., Roediger, 1990; although Roediger & Srini­ people first have the opportunity to encode a list ofwords. vas, 1993, and Schacter, 1990, have noted that these view­ Later, they are given word stems (e.g., the first three letters points may be more complementary than competing). common to a number of words) and asked to complete Although there are many well-supported, consistent them with the first word that comes to mind. Generated findings in the literature, there are also many points of words often come from the list, but research participants disagreement. Differences in research findings have been also complete stems with words that are not on the list. attributed to inadequate control ofsubject strategies (e.g., Typically, the performance ofinterest in implicit mem­ explicit memory contamination of implicit tests; Schac­ ory is priming-the proportion ofthe studied words gen­ ter, Bowers, & Booker, 1989), variation in type ofprocess­ erated at test minus a baseline, or unprimed, completion ing at study (e.g., incomplete processing of the entire rate. The baseline is usually estimated by the number of stimulus; Challis & Brodbeck, 1992), variation in test type nonstudied target words generated at test by the same re­ search participants (although Park & Shaw, 1992, advo­ cated consideration of item-based baselines along with This research was begun while R.J.S. was a postdoctoral fellow on a such subject-based baselines). An interesting problem training grant to the Georgia Institute ofTechnology, funded by the Na­ for researchers wishing to generate a list of words for tional Institute on Aging (T31 AG-OOI75). A large number ofstudents people to study is that they need to know the baseline be­ assisted in testing participants and entering, checking, and editing re­ fore they collect any data. That is, if the experimenter sponses; I am grateful to E. Chenoweth, M. Chidester, E. Deahl, C. En­ should select targets that everyone would generate by glish, R. Forbes, 1. Groark, B. Haas, K. Jones, D. Knight, D. Miller, R. Myers, S. Orey, T. Ralston, S. Salovich, M. Santore, M. Shelek, and chance to the word stem, no priming could be observed; K. Sosnow. I am also grateful to my colleagues E. Caldwell and 1.Cros­ conversely, priming might not occur for targets that no­ bie for technical assistance and encouragement. An earlier draft ofthe body would generate by chance. So the experimenter present paper was substantially improved by comments from 1.Crosbie, should select targets that have a moderate baseline. The J. Toth, M. S. Weldon, and an anonymous reviewer. Correspondence should be addressed to R. 1. Shaw, Department of Psychology, Merri­ usual procedure is to rely on luck, and collect baseline data mack College, 315 Turnpike St., North Andover, MA 01845 (e-mail: concurrently with priming data. Research participants [email protected]). are primed for halfofthe target words, and their responses 401 Copyright 1997 Psychonomic Society, Inc. 402 SHAW to the unprimed stems serve as the baseline for the other The present paper describes normative responses to a half of the targets. large collection ofthree-letter stems as a way ofprovid­ Researchers typically do not try to conduct research ing some initial information about the kinds ofwords that entirely on the basis ofluck, however. Consequently, re­ people generate ad lib., and the actual influences offre­ searchers appear to have assumed that unprimed stem quency, length, and number of alternatives. It is hoped completion rates are determined by normative word fre­ that such information will allow for the possibility ofex­ quency (based on Kucera & Francis, 1967, for example), ploring the characteristics of priming with reduced ex­ word length, and the number of available alternative com­ perimental error. pletions to the stem (usually based on counting main en­ Grafand Williams (1987) reported normative response tries in a dictionary). Graf, Mandler, and Haden (1982), information for 40 different word stems. They gave the for example, used short words (five or six letters), with 40 stems to each of 100 individuals, without giving them "at least four alternatives for completing each stem to any priming list of words. Although their data are obvi­ form a word ofthe same length" based on WebstersPocket ously useful, they are limited to only 40 stems, and this Dictionary. They noted that target words were "never the may not allow for a sufficient number oftarget words in most common completion ofthe stem" (p. 1243). Light every study. Park and Shaw (1992), for example, used 72 and Singh (1987), for example, described the same gen­ different word stems. Indeed, a quick review ofother stud­ eral characteristics (with explicit reference to Kucera & ies using word-stem completion revealed that 40 targets Francis as the basis for frequency), citing Grafet al.'s de­ is almost always too few. The present study sought to rem­ scription as the reason. Generally, however, there is little edy those limitations by collecting responses to a much consistency, even within a single researcher, in how ma­ larger set ofstems, with a larger sample ofpeople, from terials are described. two different populations. In total, 160 students responded There are two problems with collecting baseline data to each of914 word stems. These 914 stems represent a during the experiment. First, it is possible that words fairly exhaustive set ofpossible stems, using the criteria generated to nontarget stems are artificially affected by that 8 or more alternative responses have a normative the studied list. Various characteristics ofthe list may in­ word frequency (Kucera & Francis, 1967) of I or more. fluence the types of words people think of (e.g., short, concrete nouns instead ofa random selection ofall types METHOD ofwords). For example, in attempting to write down from memory a story written in an unusual style, people try to Participants mimic that style (Bartlett, 1932). It is possible that this Responses of 160 undergraduate students from psychology courses at the Georgia Institute ofTechnology (GT) and West Vir­ occurs in priming experiments, and may act on an uncon­ ginia University (WVU) were collected. Using individuals from scious level as well, influencing priming. Consequently, two institutions provided a wider variety of students than either the extent ofpriming may be underestimated by the typ­ population alone could provide. GT is a highly selective institution, ical experimental design. Ifso, true baseline information whereas WVU is a large state university with a broader admissions should come from people not exposed to an experimen­ policy. Students from both institutions participated for course credit tal list, but this is not typically done (although see Craik, in introductory psychology courses. These populations were sam­ Moscovitch, & McDowd, 1994, for an exception). pled because they are typical ofthose in implicit memory research. Individuals whose native language was not English were ex­ The second problem with collecting baseline data on­ cluded from the sample. The mean age of participants was 20.6 line is that it is likely to lead to widely varying baselines years (range, 18--40). There were 77 males and 78 females (5 peo­ within the set of words if the assumptions about length ple did not respond to the question of sex), and the majority were and frequency are not correct. It is possible that an exper­ sophomores and juniors (100 individuals). Ofthe 160 participants, imenter may inadvertently choose a relatively rare word 61 were from GT and 99 were from WVU. (by Kucera & Francis, 1967, standards) that nonetheless Materials is a common completion for its stem.