Read Writ (2012) 25:1195–1216 DOI 10.1007/s11145-011-9313-z Contribution of morphological awareness and lexical inferencing ability to L2 vocabulary knowledge and reading comprehension among advanced EFL learners: testing direct and indirect effects Dongbo Zhang · Keiko Koda Published online: 22 April 2011 © Springer Science+Business Media B.V. 2011 Abstract Within the Structural Equation Modeling framework, this study tested the direct and indirect effects of morphological awareness and lexical inferencing ability on L2 vocabulary knowledge and reading comprehension among advanced Chinese EFL readers in a university in China. Using both regular z-test and the bootstrapping (data-based resampling) methods, the study found that morphological awareness contributed to L2 vocabulary knowledge directly and indirectly through the mediation of learners’ lexical inferencing ability. It was also observed that morphological awareness made no significant unique or direct contribution to reading comprehension after adjusting for vocabulary knowledge; its indirect effects on reading comprehension, however, were significant, both through the mediation of vocabulary knowledge alone, and the multiple mediations of lexical inferencing ability and vocabulary knowledge. Keywords Morphological awareness · Vocabulary knowledge · Reading comprehension · English as a Foreign Language Introduction Studies on literacy acquisition have recently showed that morphological awareness is a significant contributor to both vocabulary knowledge and reading D. Zhang (&) Center for Research in Pedagogy and Practice, National Institute of Education, Nanyang Technological University, 1 Nanyang Walk, Singapore 637616, Singapore e-mail: dongbo.zhang@nie.edu.sg K. Koda Department of Modern Languages, Carnegie Mellon University, Baker Hall 160, 5000 Forbes Avenue, Pittsburgh, PA 15213, USA e-mail: kkoda@andrew.cmu.edu 123 1196 D. Zhang, K. Koda comprehension ability (e.g., Carlisle, 2000; Ku & Anderson, 2003; McBride-Chang et al., 2008; Nagy, Berninger, & Abbot, 2006). However, most of such research has focused on monolingual children, with limited attention given to second language (L2) reading, particularly among adult L2 learners. More importantly, the relationship of morphological awareness to vocabulary knowledge and reading comprehension in the current literature is far from clear. While first language (L1) reading research has largely confirmed direct effects of morphological awareness on these two competencies, logic suggests that the contribution of morphological awareness could also be indirect. Effects of morphological awareness on vocabulary knowledge could be partially mediated by learners’ skill to make inference about meanings of unknown words. Additionally, there is the possibility that contribution of morphological awareness to reading comprehension could be mediated by vocabulary knowledge, given the close relationship between these latter two constructs. Due to the methods used in previous studies, mostly bivariate correlation and hierarchical regression, these indirect effects of morphological awareness have not been well documented in the literature. Therefore, in this study, within the Structural Equation Modeling (SEM) framework, we tested both direct and indirect effects of morphological awareness on L2 vocabulary knowledge and reading comprehension by focusing on a group of Chinese English as a Foreign Language (EFL) readers pursuing their graduate study in a Chinese university. Morphological awareness, vocabulary knowledge, and reading comprehension Morphological awareness is the ability to reflect upon and manipulate morphemes and morphological structure of words (Carlisle, 2003; Kuo & Anderson, 2006). Because the majority of words monolingual children encounter in print school materials are multimorphemic words that are semantically transparent (Nagy & Anderson, 1984), morphological awareness has often been found to be a significant contributor to word learning and vocabulary knowledge development (e.g., Anglin, 1993; Carlisle, 2000; McBride-Chang et al., 2008; Wysocki & Jenkins, 1987). Anglin (1993) reported that from Grade 1 to 5, the number of derived words learned by English-speaking monolinguals (about 14,000) was more than three times as much as the number of root words (about 4,000) known by the same group of children. Wysocki and Jenkins (1987) found that English-speaking children’s use of morphological information facilitated their learning of new words, as they learned new words that were derivationally related to the words for which they had received prior training significantly better than control words. Carlisle (2000) also reported that derivational morphological awareness accounted for a large amount of variance in vocabulary knowledge among English-speaking third and fifth graders, suggesting strong contribution of morphological awareness to vocabulary knowledge growth. In a longitudinal study, McBride-Chang et al. (2008) tracked the development of morphological awareness and vocabulary knowledge in Chinese- and Koreanspeaking preschoolers. It was found that across language groups, children’s Time 1 compound awareness significantly predicted their vocabulary knowledge at Time 2, 123 Contribution of morphological awareness and lexical inferencing ability 1197 even when Time 1 vocabulary knowledge, phonological processing skills, and other related skills were controlled for. In addition to its importance to word learning and vocabulary knowledge acquisition, morphological awareness has also been found to be a significant contributor to reading comprehension (e.g., Carlisle, 2000; Deacon & Kirby, 2004; Ku & Anderson, 2003; Nagy et al., 2006). Deacon and Kirby (2004), for example, found that English-speaking children’s morphological awareness in Grade 2 significantly predicted their reading comprehension in Grades 3, 4, and 5, even after controlling for the influences of general intelligence, Grade 2 reading comprehension, and phonological awareness. The relationship of morphological awareness to reading comprehension becomes complex when the contribution of vocabulary knowledge to reading comprehension, widely documented in various studies in the past decades (see Nagy, 2005), is taken into consideration. Because of the interrelations between morphological awareness and vocabulary knowledge on one hand, and vocabulary knowledge and reading comprehension on the other hand, L1 reading researchers have been interested in whether morphological awareness could contribute uniquely or directly to reading comprehension over and above vocabulary knowledge. Ku and Anderson (2003), for example, showed that, after vocabulary knowledge was controlled for, morphological awareness explained a significant proportion of variance of reading comprehension, and this pattern was consistent among Grades 2, 4, and 6 English- and Chinese-speaking children. In their study on child and adolescent English readers, Nagy et al. (2006), by use of the SEM method, examined the relative contribution of morphological awareness and vocabulary knowledge to reading comprehension. It was found that morphological awareness, while contributing significantly to vocabulary knowledge, also predicted reading comprehension when the effect of vocabulary knowledge was partialed out. In addition, vocabulary knowledge also significantly predicted reading comprehension after accounting for the effect of morphological awareness. Thus, morphological awareness and vocabulary knowledge made independent contributions to reading comprehension, or their individual direct effect on reading comprehension was significant. However, that study did not address whether morphological awareness, given its close relationship with vocabulary knowledge, also contributes to reading comprehension indirectly through the mediation of vocabulary knowledge. It should be noted that evidence for the contribution of morphological awareness to vocabulary growth and reading development comes predominantly from L1 studies on monolingual children, particularly English-speaking children. Very limited research has so far been conducted in this field of inquiry on L2 learners with morphological awareness as a construct of central interest. However, the small number of studies on English L2 learners does provide some insights into the issues, and sheds some lights on the relationship of morphological skills to vocabulary knowledge and reading comprehension (e.g., Kieffer & Lesaux, 2008; Mochizuki & Aizawa, 2000; Qian, 1999; Schmitt & Meara, 1997; Wang, Cheng, & Chen, 2006; Wang, Ko, & Choi, 2009). In their study on order of English affix acquisition, Mochizuki and Aizawa (2000) found that Japanese high-school and university EFL learners’ knowledge of 123 1198 D. Zhang, K. Koda English prefixes and suffixes was significantly positively correlated with their vocabulary size. Also focusing on Japanese EFL learners, Schmitt and Meara (1997) reported positive correlations between learners’ knowledge of English derivational affixes and word association ability (i.e., the ability to find related words for prompt verbs, such as refuse and accept for deny). Studies also revealed that word morphology is an important knowledge source that adult EFL learners use to make informed guesses about meanings of unknown words during textual reading (e.g., Paribakht & Wesche, 1999). All these research findings point to the importance of morphological knowledge in word learning and vocabulary development among English L2 learners. Few studies have examined issues related to the contribution of morphological awareness to reading comprehension among English L2 readers with learners’ vocabulary knowledge also considered. Kieffer and Lesaux (2008) tracked the development of the three competencies and other reading-related skills among a group of Spanish-speaking English as a Second Language (ESL) readers from Grade 4 to 5 in the United States. Hierarchical regression analysis showed that there was significant independent contribution of learners’ morphological awareness and vocabulary knowledge to reading comprehension in Grade 5. Similarly, Wang and associates (Wang et al., 2006, 2009) found Chinese and Korean immigrant children’s ESL morphological awareness significantly predicted their English reading comprehension, after the effects of English oral vocabulary knowledge as well as phonemic awareness were accounted for. However, Qian’s (1999) study did not reveal a unique contribution of morphological knowledge to reading comprehension over and above vocabulary knowledge among Chinese and Korean ESL readers in Canadian universities. Overall, a comprehensive picture is hard to obtain from the existing body of research about the contribution of morphological awareness to L2 vocabulary knowledge and reading comprehension. Further studies are needed to model the complex relations between these variables within an integrated framework. Lexical inferencing, word learning, and textual comprehension Lexical inferencing means “making informed guesses as to the meaning of a word, in light of all available linguistic cues in combination with the learner’s general knowledge of the world, her awareness of context and her relevant linguistic knowledge” (Haastrup, 1991, p. 40). Thus, in addition to contextual clues, learners’ ability to use “linguistic cues” or morphological information in words also serves a foundation for successful inference of meaning. Anglin (1993, p. 5) called this process of learning new word as “morphological problem-solving.” That is, children use their morphological knowledge to decompose complex words into their morphemic constituents and infer meaning of novel vocabulary items based on the meanings of these constituents. Because the focus of the present study was on morphological processes, we chose to look only at morphology-based lexical inferencing ability. In other words, lexical inferencing ability was operationally defined in the present study as learners’ ability to infer meanings of unknown 123 Contribution of morphological awareness and lexical inferencing ability 1199 multimorphemic words by drawing upon intra-word morphological cues. While contextual guessing has also been confirmed in previous research as critical to lexical inferencing, it was not considered in the present study. Previously, various studies on L2 reading have revealed the importance of morphology-based lexical inferencing in word learning and vocabulary knowledge development. Gu and Johnson (1996) found that Chinese college EFL readers’ rating on their use of the strategy of paying attention to word formation was significantly and positively correlated with their vocabulary size. In Qian’s (2004) study, Korean and Chinese ESL readers in Canadian universities reported that they often used intra-word clues, in addition to contextual guessing, when they dealt with unknown words while reading. Paribakht and Wesche’s (1999) introspective study also found word morphology to be one of the most frequent knowledge sources learners used in performing lexical inferencing to unlock meanings of new words. Given its importance to instantaneous resolution of vocabulary gaps during reading or “on the spot vocabulary learning” (Nagy, 2007, p. 64), lexical inferencing, while contributing to word learning, should also enhance textual comprehension. However, current literature has not yet produced a clear picture on the relationship of lexical inferencing skill to reading comprehension ability. To begin with, lexical inferencing ability as defined in the present study was often measured as a particular facet of morphological awareness in previous research where the relationship between morphological awareness and reading comprehension was examined (e.g., Ku & Anderson, 2003), possibly due to different interests of researchers. Because this ability was often calculated with other morphological skills to produce a holistic score to index morphological awareness (e.g., factor scores in Ku & Anderson, 2003), its relationship with those morphological skills, and, more importantly, its specific contribution to reading comprehension have not been known. So far, there has been very little, if any, empirical research that has directly tested the relationship of lexical inferencing ability to reading comprehension, particularly when morphological awareness and vocabulary knowledge were also considered as separate variables. Consequently, it remains unclear whether lexical inferencing skill would contribute directly or uniquely to reading comprehension, and whether there would also exist any indirect influence of lexical inferencing on reading comprehension through the mediation of vocabulary knowledge, given the contribution of lexical inferencing to vocabulary development on one hand, and that of vocabulary knowledge to reading comprehension on the other hand. In order to have a clear understanding of the relationship between morphological awareness, lexical inferencing skill, vocabulary knowledge, and reading comprehension, these direct and indirect effects have to be disentangled. The present study The above review of the literature suggests that how morphological awareness contributes to vocabulary knowledge, and subsequently, reading comprehension, is far from clear. While previous studies have provided some insights into this issue, a few important questions are yet to be answered. First, previous research has been 123 1200 D. Zhang, K. Koda largely focused on monolingual children, and in the few cases where English L2 learners were involved, findings are sometimes are mixed. To better understand how morphological awareness functions in vocabulary knowledge and reading development, further studies on L2 learners are thus necessitated, particularly with the intermediate processes (e.g., lexical inferencing ability) considered. Second, because of its reliance on bivariate correlations and conventional regression analyses, previous research has not well explored the intermediate processes, and the evidence that supports the direct as well as indirect effects of morphological awareness is by no means conclusive. Therefore, focusing on adult EFL learners in China, and with the use of the SEM as well as the data-based resampling methods, this study aimed to unpack the complex relationship of morphological awareness and lexical inferencing to L2 vocabulary knowledge and reading comprehension. Specifically, the study aimed to address the following two questions: 1. 2. Do Chinese university EFL learners’ morphological awareness and lexical inferencing ability have independent or direct effect on vocabulary knowledge? Does lexical inferencing mediate the effect of morphological awareness on vocabulary knowledge? Do learners’ morphological awareness, lexical inferencing ability, and vocabulary knowledge directly contribute to their reading comprehension, adjusting for the effects of any two of these variables? Are there any significant indirect effects of morphological awareness on reading comprehension through the mediation of learners’ vocabulary knowledge and/or lexical inferencing ability? Method Participants Participants were 130 adult EFL learners working on their engineering master’s degrees in a university in Shanghai, China. They consisted of 33 females and 97 males, with an average age of 23.85 (SD = 1.19), from four Comprehensive English classes. Forty-two of them had passed Band 4 of the College English Test, and 88 had passed both Band 4 and Band 6 of the test (see Jin & Yang, 2006 for information about this English proficiency test widely administered to college students in China). All participants had also passed The Graduate School Entrance English Examination (He, 2010), and were taking the Comprehensive English course to fulfill the foreign language requirement for the first semester of their graduate study. Each week they met with their teachers in classroom for three 1.5-h sessions to develop their reading and writing as well as oral communication skills. Most of the participants started formal EFL learning from Grade 7 (N = 109), with a small number from a mixture of grades (four from Grade 3, three from Grade 4, 11 from Grade 5, and three from Grade 6). At the time of data collection, those who started English learning from Grade 7 had received about 8 years of formal 123 Contribution of morphological awareness and lexical inferencing ability 1201 EFL instruction. The participants’ average year of formal EFL education was roughly 8.88 (SD = .96). Instruments Morphological awareness Morphological awareness in this study was defined as the ability to analyze multimorphemic English words appropriately into their morphological units and correctly identify the root on which the meaning of each target word was based. To measure this competence, a task with 30 test items and two practice items was constructed. All the target words were derivatives, each of which was followed by three meaningful words with strings of letters being part of the target word (with minor difference of one or two letters in a few test items). For example, a target derivational word reforestation was followed by three words station, forest, and rest. To display their morphological awareness, learners should choose forest as the correct answer. Some considerations were taken to increase the validity of the test. Because the test aimed to measure morphological awareness rather than learners’ existing knowledge of the target words, only low-frequency (about five per million words, SD = 9.9; Kucera & Francis, 1967) derivational words unfamiliar to the participants were chosen as the target words. In addition, the roots and the distracter words had to be familiar to learners. To address these concerns, an initial version of the test was constructed after consulting the participants’ two English teachers, both of whom had had about 8 years in teaching graduate level EFL courses in the university where the participants were selected. To ensure that the participants were unfamiliar with the target words, 20 students who were also taking the Comprehensive English course but did not participate in the current study were asked to rate their familiarity with the pilot test items on a 5-point Likert scale, from very familiar (5) to very unfamiliar (1). Those items, very small in number, that received an average rating higher than 3 were dropped and replaced by more appropriate items. Finally, the affixes in the target words and the 90 high-frequency word choices (about 52 per million words, SD = 112.6; Kucera & Francis, 1967) in the final version of the task had appeared in the textbooks previously used by the participants, and were reported by their teachers as familiar to the participants. Participants were given 5 min to complete the test. They received one point for each correct choice. The maximum score was 30. The reliability (Cronbach’s α) was .75. Lexical inferencing ability As mentioned earlier, lexical inferencing was narrowly defined as making appropriate predictions of meanings for unknown morphologically complex words based on intra-word morphological cues. To make successful meaning inferences, learners not only need to be attentive to the functions and meanings of affixes, but should be able to synthesize or integrate structural (morphological structure) and 123 1202 D. Zhang, K. Koda semantic (morpheme meaning) information of each target word. The lexical inferencing test consisted of 30 low-frequency (about two per million words, SD = 2.18; Kucera & Francis, 1967) derivational words followed by four meaning explanations in the learners’ L1 Chinese. We took similar steps to ensure that the target words were unknown to the participants to minimize potential confound from learners’ existing knowledge of these words as we did in constructing the morphological awareness test. The meaning explanations were constructed focusing primarily on the suffixes in the target words. The participants had to know that a suffix provides clues to the part-of-speech as well as overall meaning of an unfamiliar derived word. For example, the word meritorious was followed by four choices: 给予奖励 (to reward), 功勋卓著的人 (people with outstanding contributions), 值得称赞的 (praiseworthy), 杰出的贡献 (outstanding contributions). Learners should choose the third answer if they know that the suffix -ous stands for an adjective that describes the quality of things and subsequently combine this information with the meaning of merit. Participants were given 10 min to complete this test. They received one point for each correct choice. The maximum score was 30. The reliability was .72. Vocabulary knowledge Both size and depth of vocabulary knowledge were measured. Vocabulary size was measured with an improved version of Nation’s (1990) Vocabulary Levels Test (VLT) designed by Schmitt, Schmitt, and Clapham (2001). This new version consisted of five parts, representing five levels of word frequency in English: 2,000 words (e.g., electric, independent), 3,000 words (e.g., champion, administration), 5,000 words (e.g., mortgage, nomination), university word level (e.g., infrastructure, analogous), and 10,000 words (e.g., nocturnal, expulsion). Each part of the test was comprised of 10 sets of six randomly sampled words (12 sets for the university word level). Participants were required to select three words from each set to match the three meaning explanations provided (e.g., target words: analogous, objective, potential, predominant, reluctant, subsequent; meaning choices: happening after, most important, not influenced by personal opinion). Because the participants for the current study were EFL learners with relatively high English proficiency, the 2,000 words level was considered too simple to differentiate between learners and show individual differences; therefore, it was dropped, and only the other four parts were used. Eventually, there were 126 target words in the final version of the test, among which about one third were derived words. Note that the VLT, as a standardized vocabulary size test, primarily involves retrieval of word meanings from the mental lexicon without explicitly requiring online morphological analysis and meaning inference, distinguishing it from the lexical inferencing task as delineated above. Participants received one point for correctly choosing a meaning explanation for a word. The maximum score was 126. Participants were given half an hour to complete the test. The reliability was .89. Vocabulary depth was measured with Read’s (1993, 1998) Word Associates Test (WAT). In the test, a word associates item consisted of a target adjective 123 Contribution of morphological awareness and lexical inferencing ability 1203 (e.g., beautiful) followed by eight words located in two different columns, each containing four words (e.g., enjoyable, expensive, free, loud; education, face, music, weather). Among these eight words, four words were associates of the target word. The four words in the left column were adjectives; the associates in this group were either synonyms of the target word or represented one aspect of the various meanings of the target word. The four words on the right column were all nouns; the associates in this group could collocate with the stimulus word. The test consisted of 40 items. To minimize the effect of blind guessing, which had been identified to constantly influence test-takers’ performance on the WAT (Read, 1998), in the present study participants were not informed of the exact number of correct answers; instead, they were encouraged to choose as many appropriate answers as possible. Students were also told that they would receive one point for both choosing a correct word and not choosing an incorrect word; and that they would receive nothing for either choosing an incorrect word or not choosing a correct word. This criterion, different from the original scoring criterion that only valued the choices of four correct words, was used in this study to grade learners’ answers. Hence, the maximum score for an item was eight, with the four associates being chosen and the four distracters not chosen; the minimum score for an item was zero, with the four associates not chosen and the four distracters chosen. A person who did not choose any word for an item also received no points. The maximum score was 320. Students were given 40 min to complete this test. The reliability was .76. Reading comprehension Five subskills related to the underlying processes of reading comprehension were measured with five multiple-choice questions for each of the six passages we selected. The word supply question required the participants to find an appropriate word from four high-frequency word choices to fill in the blank in each text based on local comprehension. The conjunction question asked the participants to choose a conjunction word from among four choices that best represented the relationship between two clauses. The co-reference question required the participants to identify a pronoun’s antecedent based on surrounding discourse information. The textual inference question tested learners’ ability to generate links between different parts of a text and fill in the missing details. The gist question tested learners’ competence to summarize main ideas of the passages. All questions were followed with four choices, and the participants were asked to choose the best one to answer each question. The six passages included four expository texts and two narrative texts, with an average length of 544 words (SD = 115). Four were selected from retired test papers of the Test for English Majors-Band 8 (TEM-8) (TEM-8 Development Committee, 2003) and two from retired Graduate Record Examination (GRE) tests. Most of the original items were replaced by the new ones specifically designed for this study; some items measuring textual inference and gist that were considered appropriate for the present study were kept. Participants were given 10–15 min to complete each 123 1204 D. Zhang, K. Koda passage, depending on the text length. The maximum score was 30. The reliability was .80. Data collection and analysis procedures All the tests were printed on papers and administered to the participants in a whole class format by their teachers in English classes. To avoid potential confound of learners’ decoding ability on their performance on the outcome measures, the English teachers were around to answer questions related to the pronunciations of the words in the tests. The morphological awareness test and the lexical inferencing test were first administered in one session, followed by two vocabulary knowledge tests in two sessions, and finally the reading comprehension test with six passages in six sessions. The whole process of data collection lasted about 4 weeks. Seventeen participants who did not complete all the tests were removed from the dataset, and 113 cases with intact data were kept for the final analyses. In analyzing the data, we first used SPSS 16.0 to calculate the means and standard deviations of all observed variables, and the correlations between them. The SPSS data were then retrieved in EQS 6.1 for SEM analysis (Bentler, 2005). All the following analyses of the measurement and the structural models were conducted using EQS 6.1 with the two-step modeling approach (Kline, 2005). This meant that we first used Confirmatory Factor Analysis to find an acceptable measurement model, and then imposed this measurement model on the structural model to identify the path coefficients and analyze the direct and indirect effects of the variables concerned. Since χ2 value is sensitive to sample size and is often significant with a large sample size, fit indices are adopted in covariance structure analysis to examine goodness of model fit. Different cutoff values have been proposed for different indices. Hu and Bentler (1999) suggested that Comparative Fit Index (CFI) larger than .95, and Root Mean Square Error of Approximation (RMSEA) smaller than .06 indicate a very good model. These two indices were used in the present study to make judgment on goodness of model fit. In testing the indirect effects of morphological awareness on vocabulary knowledge and reading comprehension, and those of lexical inferencing on reading comprehension, we were also interested in whether lexical inferencing and vocabulary knowledge served as mediators (M) for the effects of independent variables (IV) on dependent variables (DV), as the indirect effect of an IV on a DV does not necessarily suggest existence of a mediation effect. To establish mediation, some requirements have to be met. Baron and Kenny (1986) pointed out four conditions: (1) an IV significantly predicts a DV; (2) an IV significantly predicts a M; (3) a M significantly predicts a DV; and (4) an IV does not predict a DV adjusting for a However, Kenny, Kashy, and Bolger (1998) questioned whether all the above conditions have to be met. For example, condition 4 is only for establishing complete mediation. For partial mediation, it does not need to be a requirement. Many data analysts also argued against the necessity of condition 1, suggesting that that condition may already be implied if conditions 2 and 3 are met. In other words, the direct effect of an IV on a DV may not be observed if multiple 123 Contribution of morphological awareness and lexical inferencing ability 1205 mediation effects are present and cancel each other out (Wu & Zumbo, 2008). Based on these arguments, we used only conditions 2 and 3 as guidelines in testing whether lexical inferencing and vocabulary knowledge served as mediators of the effects of morphological awareness. Results Bivariate correlations Table 1 shows the means and standard deviations of all observed variables and the zero-order correlations between them. Most of the observed variables were correlated significantly and positively. Morphological awareness was significantly correlated with lexical inferencing ability, vocabulary size, and vocabulary depth. The two vocabulary knowledge measures also had good positive correlation (r = .546, p \ .01). Morphological awareness overall did not have significant correlations with reading comprehension. Lexical inferencing ability was closely related to two comprehension subskills: conjunction (r = .218, p \ .05) and word supply (r = .268, p \ .01). Vocabulary size had significant correlations with conjunction (r = .197, p \ .05), word supply (r = .213, p \ .05), and textual inference (r = .190, p \ .05). However, no significant relationship was found between vocabulary size and co-reference resolution and gist-making subskills. Vocabulary depth was correlated significantly with only two comprehension subskills, gist (r = .191, p \ .05) and conjunction (r = .247, p \ .01). Table 1 Descriptive statistics of and zero-order correlations between morphological awareness, lexical inferencing, vocabulary knowledge, and reading comprehension subskills 1 2 3 4 5 6 7 8 9 1 MA – 2 LEXINF .439*** 3 VOCSIZ .429*** .343*** – 4 VOCDEP .326*** .364*** .546*** 5 READCOF .152 .109 .061 .104 – 6 READGST .083 .126 .058 .191* .351*** 7 READCON .160 .218* .197* .247** .286** .039 – 8 READTEX −.031 .000 .213* .121 .191* .053 .127 – 9 READWS .223* .268** .190* .164 .249** .170 .253** .315*** – Mean 24.65 17.98 85.27 230.20 3.51 3.34 3.37 2.62 2.74 SD 2.36 4.41 11.85 17.27 1.37 1.17 1.48 1.33 1.38 MSP 30 30 126 320 6 6 6 6 6 – – – N = 113. MA morphological awareness, LEXINF lexical inferencing ability, VOCSIZ vocabulary size, VOCDEP vocabulary depth, READCOF co-reference, READGST gist, READCON conjunction, READTEX textual inference, READWS word supply, MSP maximum score possible * p \ .05, ** p \ .01, *** p \ .001 123 1206 D. Zhang, K. Koda Measurement models As indicated earlier, we took the two-step modeling approach to SEM analysis by first establishing an acceptable measurement model and then testing the structural models. In the conceptual measurement model, vocabulary size and depth were to load on the latent factor of vocabulary knowledge, and all the reading subskills on that of reading comprehension. In testing the model, we fixed the factor variances at 1.0, and allowed them to covary. EQS output showed Mardia’s coefficient = .052, with normalized estimate = .025, indicating non-significant multivariate kurtosis. Therefore, no adjustment of non-normality was needed, and regular χ2 could be used for testing model fit and for model comparisons. The initial model showed χ2 (13, N = 113) = 20.418, p = .085 (CFI = .915; RMSEA = .071, Confidence Interval or CI: .000–.101). Based on the cut-off values suggested by Hu and Bentler (1999), the model did not show very good fit. The Lagrange Multiplier Test result suggested that covarying the residuals of co-reference and gist would significantly improve the model fit. Considering that successful resolution of co-referential relationship in reading comprehension helps increase overall textual understanding, and vice versa, we allowed the error terms of the two observed variables to covary. The optimized model showed χ2 (12, N = 113) = 12.898, p = .377 (CFI = .990, RMSEA = .026, CI: .000–.092), with very good model fit and significant improvement on the initial model (Δχ2 [1] = 7.52, p = .006). Thus, it was accepted as the final measurement model. Table 2 provides further details of the standardized parameter estimates for the measurement model. Both vocabulary size (β = .748, p \ .001) and depth (β = .730, p \ .001) were well loaded on the latent factor of vocabulary knowledge. The factor explained about 56 and 53.3% of the variances in vocabulary size and depth, respectively. The subskills of reading comprehension were generally well loaded on the latent factor as well (all ps \ .001 except for gist). The factor loading of word supply was β = .593, with about 35.2% of its variance explained by the factor. For conjunction, β = .466, with about 21.9% of variance explained by the factor. For co-reference and textual inference, β = .440, and β = .446, respectively. Reading comprehension explained about 19.4 and 19.9% of the variance in these two observed variables, respectively. The only observed variable that was not well loaded on the factor was gist, β = .218, p = .083. Finally, the two factors were also significantly correlated, r = .454, p \ .001. Testing direct and indirect effects in the baseline structural model The above final measurement model was imposed upon a structural model for testing goodness of model fit. Based on the literature review, we came up with a conceptual structural model as shown in Fig. 1, where reading comprehension was predicted by three variables: morphological awareness, lexical inferencing ability, and the latent variable of vocabulary knowledge. In addition, vocabulary knowledge was predicted by both morphological awareness and lexical inferencing, and lexical inferencing was predicted by morphological awareness. In testing the structural 123 Contribution of morphological awareness and lexical inferencing ability 1207 Table 2 Parameter estimates of the optimized measurement model for confirmatory factor analysis of vocabulary knowledge and reading comprehension Paths β z p R2 VOCSIZ ← VOCAB .748 4.935 \.001 .560 VOCDEP ← VOCAB .730 4.887 \.001 .533 READWS ← READ .593 4.715 \.001 .352 READCON ← READ .466 3.853 \.001 .219 READCOF ← READ .440 3.625 \.001 .194 READTEX ← READ .446 3.678 \.001 .199 READGST ← READ .218 1.733 .083 .047 a E8 (READCOF) ↔ E10 (READGST) .407 2.608 .009 – VOCAB ↔ READ .454 3.933 \.001 – N = 113. VOCSIZ vocabulary size, VOCDEP vocabulary depth, VOCAB latent factor of vocabulary knowledge, READWS word supply, READCON conjunction, READCOF co-reference, READTEX textual inference, READGST gist, READ latent factor of reading comprehension a Error terms of READCOF and READGST were allowed to covary Fig. 1 The conceptual structural model for the relationships between morphological awareness, lexical inferencing, vocabulary knowledge, and reading comprehension model, factor loadings of vocabulary size and word supply were fixed at 1.0. EQS output showed Mardia’s coefficient = 1.75, with normalized estimate = .662, indicating, again, non-significant multivariate kurtosis of the structural model. No adjustment of non-normality was thus needed and regular χ2 could be used for testing model fit and for model comparisons. The baseline structural model showed χ2 (22, N = 113) = 25.417, p = .278 (CFI = .976; RMSEA = .037, CI: .000–.090) with very good fit. Table 3 shows the standardized parameter estimates of that model, with decomposed effects (direct and indirect effects) and total effects. Morphological awareness significantly predicted lexical inferencing, β = .439, p \ .001, and explained about 19.3% of its variance. Both morphological awareness and lexical inferencing made significant independent contribution to vocabulary knowledge (34.1% variance explained) after controlling for each other’s effect, β = .386, p \ .001, and β = .300, p = .006, respectively. Morphological awareness, lexical inferencing, and vocabulary knowledge together explained about 22.1% of the variance of reading comprehension. 123 1208 D. Zhang, K. Koda Table 3 Parameter estimates of the baseline structural model for the relationships between morphological awareness, lexical inferencing, vocabulary knowledge, and reading comprehension with direct, indirect, and total effects Predictors Direct effect Indirect effect 2 β z p .439 5.173 \.001 R LEXINF Total effect β z p β z p – – – .439 5.173 \.001 .193 ← MA VOCAB .341 ← MA .386 3.452 \.001 ← LEXINF .300 2.723 .006 READ .132 2.409 .016 .518 4.814 \.001 – – – .300 2.723 .006 .221 ← MA .042 .277 .782 .250 2.354 .019 .292 2.292 .022 ← LEXINF .172 1.193 .233 .101 1.480 .139 .274 1.993 .046 ← VOCAB .337 1.714 .087 – – – .337 1.714 .087 N = 113. MA morphological awareness, LEXINF lexical inferencing ability, VOCAB latent factor of vocabulary knowledge, READ latent factor of reading comprehension Morphological awareness seemed to have the least unique contribution to reading comprehension after controlling for lexical inferencing and vocabulary knowledge, β = .042, p = .782; the independent contribution of lexical inferencing was a little higher, β = .172, p = .233. Vocabulary knowledge had the largest independent contribution to reading comprehension, β = .337, p = .087. However, no one variable had any significant unique or direct effect on reading comprehension after adjusting for the other two variables. As Table 3 shows, in the baseline structural model, there was also a significant indirect effect of morphological awareness, through lexical inferencing ability, on vocabulary knowledge (γ21β32), β = .132, p = .016. Both direct and indirect effects being significant, the total effects of morphological awareness on vocabulary knowledge were significant too, β = .518, p \ .001. There were significant overall indirect effects of morphological awareness on reading comprehension (γ31β13 + γ21β12 + γ21β32 β13), β = .250, p = .019, and the total effects were also significant, β = .292, p = .022. Optimizing the structural model and testing effects in the new model The Wald Test result suggested that removing the paths from morphological awareness and lexical inferencing to reading comprehension would make the structural model more parsimonious, but bring no significant change to the goodness of model fit. Because the indirect effect of morphological awareness on reading comprehension had to be tested adjusting for its direct effect, we did not remove the path between these two variables, even though the direct effect was not significantly different from zero. Instead, we dropped the path from lexical inferencing to reading comprehension to improve the parsimony of the structural model. The optimized model showed χ2 (23, N = 113) = 26.763, p = .266 (CFI = .974; RMSEA = .038; 123 Contribution of morphological awareness and lexical inferencing ability 1209 Table 4 Parameter estimates of the optimized structural model for the relationships between morphological awareness, lexical inferencing, vocabulary knowledge, and reading comprehension with direct, indirect, and total effects Predictors Direct effects Indirect effects 2 β z p .439 5.173 \.001 R LEXINF Total effects β z p β z p – – – .439 5.173 \.001 .193 ← MA VOCAB .354 ← MA .379 3.387 \.001 ← LEXINF .321 2.910 .004 READa .141 2.536 .011 .520 4.783 \.001 – – – .321 2.910 .004 .223 ← MA .060 .394 .694 .228 2.113 .035 .287 2.232 .026 ← VOCAB .438 2.253 .024 – – – .438 2.253 .024 N = 113. MA morphological awareness, LEXINF lexical inferencing ability, VOCAB latent factor of vocabulary knowledge, READ latent factor of reading comprehension a The path from LEXINF to READ was removed CI: .000–.090) with good fit but no significant difference from the baseline model, Δχ2 (1) = 1.346, p = .246. Therefore, this more parsimonious model was accepted as the final structural model for further testing of direct, indirect, and total effects of the predictors. Table 4 shows minimal changes in the final structural model from the baseline model (see Table 3) with regard to the effects of morphological awareness and lexical inferencing on vocabulary knowledge. The direct effects of the two variables on vocabulary knowledge remained similarly significant, so were the indirect and total effects of morphological awareness on vocabulary knowledge. However, with the removal of the path from lexical inferencing to reading comprehension, the unique contribution of vocabulary knowledge to reading comprehension, adjusting for morphological awareness, became significant, β = .438, p = .024, but not vice versa (β = .060, p = .694). This suggested the more powerful influence of vocabulary knowledge on reading comprehension. The indirect effects (β = .228, p = .035) and total effects (β = .287, p = .026) of morphological awareness on reading comprehension remained similarly significant as in the baseline model. It should be pointed out that there were two limitations in the above analyses of indirect effects. First, all indirect effects were examined using z-tests (Sobel, 1982), which require a normal sampling distribution. However, as an indirect effect is multiplicative (e.g., the indirect effect of morphological awareness on vocabulary knowledge γ21 9 β32), its sampling distribution is often unknown and seldom normally distributed. Therefore, the indirect effects reported above could be biased. Shrout and Bolger (2002) proposed that bootstrapping, a data-based simulation method for statistical inference, be used for testing indirect effects. The bootstrapping method uses empirical sample data to generate a certain number of bootstrap samples (typically 1000) through random sampling with replacement. Each bootstrap sample is then analyzed to estimate a parameter. As the method uses 123 1210 D. Zhang, K. Koda confidence interval to determine whether a statistic is significantly different from zero, it does not have the assumption of normality of sampling distribution. Cheung and Lau (2008) found that the bootstrapping method performed the best, compared to a number of other methods, in estimating indirect effects. Second, EQS, like other SEM programs, only reports total indirect effects. When there are multiple indirect effects of an IV on a DV through different paths, the indirect effects via each individual path and their relative contribution to the overall indirect effects are not known. For example, in our final structural model (see Table 4), morphological awareness indirectly contributed to reading comprehension via two paths, γ31β13 and γ21β32β13. Reported in Table 4 were the total indirect effects of these two paths. Therefore, it was not known whether morphological awareness indirectly contributed to reading comprehension through vocabulary knowledge alone (γ31β13), or through both lexical inferencing and vocabulary knowledge (γ21β32β13). The bootstrapping method provided a good way to estimate these component indirect effects. Given the above limitations, 1,000 bootstrap samples were generated from the original empirical data. We requested percentile scores of 2.5% and 97.5% for computing 95% CI to examine the frequency distribution of indirect effects parameters to be estimated. As shown in Table 5, the indirect effect of morphological awareness on vocabulary knowledge via lexical inferencing (γ21β32) was significant (95% CI: .1105–1.1410). The overall indirect effect of morphological awareness on reading comprehension (γ31β13 + γ21β32β13) was significant (95% CI: .0187–.1820). The two component indirect effects of morphological awareness on reading comprehension, one via vocabulary knowledge (γ31β13; 95% CI: .0126–.1439), and the other via both lexical inferencing and vocabulary knowledge (γ21β32β13; 95% CI: .0025–.0563), were both significant too. To answer the research questions as to whether lexical inferencing and vocabulary knowledge serve as mediators for the indirect effects, we went back to conditions 2 and 3 given by Baron and Kenny (1986). We found lexical inferencing ability significantly mediated the indirect effect of morphological awareness on vocabulary knowledge (γ21β32), as it was significantly predicted by morphological awareness (γ21 ≠ 0; condition 2), and it significantly predicted vocabulary knowledge adjusting for morphological awareness (β32 ≠ 0; condition 3). In addition, the indirect effects of morphological awareness on reading comprehension were mediated by vocabulary Table 5 95% confidence intervals for examining indirect effects of morphological awareness on vocabulary knowledge and reading comprehension Indirect effects 95% confidence intervals MA → LEXINF → VOCAB (γ21β32) .1105–1.1410 MA → VOCAB → READ (γ31β13) .0126–.1439 MA → LEXINF → VOCAB → READ (γ21β32β13) .0025–.0563 MA → READ (γ31β13 + γ21β32β13) .0187–.1820 MA morphological awareness, LEXINF lexical inferencing ability, VOCAB latent factor of vocabulary knowledge, READ latent factor of reading comprehension; All indirect effects were significant 123 Contribution of morphological awareness and lexical inferencing ability 1211 knowledge alone (γ31β13), as morphological awareness significantly predicted vocabulary knowledge adjusting for lexical inferencing (γ31 ≠ 0; condition 2), and vocabulary knowledge significantly predicted reading comprehension adjusting for morphological awareness (β13 ≠ 0; condition 3). As γ21 ≠ 0, β32 ≠ 0, and β13 ≠ 0, vocabulary knowledge and lexical inferencing ability also jointly mediated the indirect effects of morphological awareness on reading comprehension. Discussion Previous studies on monolingual children (e.g., Anglin, 1993; McBride-Chang et al., 2008) have shown that awareness of English morphological structure plays an important role in vocabulary knowledge acquisition. Through “morphological problem solving” (Anglin, 1993, p. 5), children learn new complex words by inferring their meanings (Nagy & Anderson, 1984). Studies on adult EFL/ESL learners have also confirmed that knowledge of affixes, lexical inferencing skill, and vocabulary knowledge are closely related (e.g., Mochizuki & Aizawa, 2000; Paribakht & Wesche, 1999; Schmitt & Meara, 1997). These findings from previous studies imply that the effect of morphological awareness on vocabulary knowledge, at least some part of it, should be achieved through learners’ lexical inferencing skill. No studies, however, have directly examined this issue, particularly among L2 learners. Empirically, our study, focused on adult Chinese EFL readers, supports the existence of a partial mediation effect of lexical inferencing skill. In other words, some effect of morphological awareness on EFL vocabulary knowledge was realized via learners’ skill to integrate structural (derivational morphology) and semantic (morpheme meaning) information to infer meanings of unknown complex words. This mediation effect did not come as a surprise, given that our morphological awareness task measured learners’ root identification or morphological segmentation ability and this ability was critical to their subsequent use of intra-word clues for meaning inferencing. As a result, those learners who possessed better morphological awareness tended to learn words better, and in turn, held a larger vocabulary. This finding also accords with Nagy’s (2007) observation, situated in English monolingual children’s vocabulary acquisition, that at least part of the relationship between word structure knowledge and vocabulary acquisition could be attributed to the effect of morphological awareness on word learning ability. We also found that morphological awareness, in addition to indirectly contributing to vocabulary knowledge via lexical inferencing ability, also had a unique or direct effect on vocabulary knowledge. Existence of this direct effect is not surprising, because learners’ insights into stems, affixes, and word structure, while helping with meaning inference, also strengthen their representation of complex words in the mental lexicon and enhance retention of vocabulary items (Sandra, 1994). Our second set of research questions asked about the contribution of morphological awareness to L2 reading comprehension, in conjunction with lexical inferencing ability and vocabulary knowledge. According to our final model 123 1212 D. Zhang, K. Koda (see Table 4) and bootstrapping-based parameter estimates, morphological awareness had significant indirect effects on reading comprehension, through the mediation of vocabulary knowledge alone, and the multiple mediations of both lexical inferencing and vocabulary knowledge. To put it simply, indirect effects of morphological awareness on L2 reading comprehension were largely realized via vocabulary knowledge. This finding seems to accord with those of previous L2 studies that have documented a close relationship between morphological and vocabulary knowledge on one hand (Mochizuki & Aizawa, 2000; Schmitt & Meara, 1997), and that between vocabulary knowledge and reading comprehension on the other hand (Laufer, 1992). Thus, the shared variance between L2 morphological awareness and reading comprehension could be attributed to their joint variance with vocabulary knowledge. Interestingly, however, results of the current study did not support our conceptual structural model, which proposed that each one of the three variables, morphological awareness, lexical inferencing ability, and vocabulary knowledge, uniquely contributed to reading comprehension after the other two variables were controlled for. In other words, no one variable was found to have any significant independent contribution to learners’ reading comprehension performance. This seems to contradict previous L1 and L2 research findings and observations about the unique contribution of morphological awareness to reading comprehension over and above vocabulary knowledge. Both Ku and Anderson (2003) and Nagy et al. (2006), for example, found that monolingual English-speaking children’s morphological awareness significantly predicted their reading comprehension even after partialing out the influence of vocabulary knowledge. Among studies on English L2 learners, Kieffer and Lesaux (2008) also found Grade 5 Spanish-speaking ESL readers’ morphological awareness and vocabulary knowledge had significant independent contribution to reading comprehension. According to Nagy (2007), the unique contribution of morphological awareness to reading comprehension over and above vocabulary knowledge could be attributed to the mediating effect of lexical inferencing ability. That is to say, morphological awareness facilitates meaning inferencing of complex words during reading; this in turn helps learners resolve vocabulary gaps in reading and lead to better comprehension. In addition, learners can use the syntactic signals provided by suffixes in derived words to help parse complex sentences, which could also contribute to comprehension. Finally, morphological awareness may also contribute to comprehension via its effects on fluency of word decoding. Why were the functional relations between morphological awareness and reading comprehension not found in the current study as it has been discussed by reading researchers and documented in previous research? We conjecture that the disagreements might be attributed to the following reasons. To begin with, previous studies that supported the independent contribution of morphological awareness to reading comprehension only controlled for learners’ vocabulary size, while the depth dimension was not measured as a part of vocabulary knowledge (e.g., Ku & Anderson, 2003; Nagy et al., 2006). This study, however, employed both size and depth as indicators of vocabulary knowledge, which could have strengthened the contribution of vocabulary knowledge to reading comprehension and led to limited 123 Contribution of morphological awareness and lexical inferencing ability 1213 direct effect of morphological awareness on reading comprehension. In this regard, the current study corroborates Qian (1999), which found morphological knowledge did not contribute significantly to reading comprehension among adult ESL learners in Canadian universities, after both vocabulary size and depth were controlled for. The diminishing unique contribution of morphological awareness in this study might also be due to the measurement of morphological awareness. The morphological awareness task in the current study only dealt with learners’ ability to segment unfamiliar English derived words into morphological constituents and identify the base forms of the target words. There is a possibility that this morphological insight might have limited direct contribution to reading comprehension. Its effect on reading comprehension might have to be mediated by meaning inferencing ability and vocabulary knowledge. This was partly reflected in the low correlations of our morphological awareness measure with the five comprehension subskills. In contrast to our study, previous studies that have found unique contribution of morphological awareness to reading comprehension often employed a number of measures tapping different levels of morphological awareness (e.g., Kieffer & Lesaux, 2008; Ku & Anderson, 2003; Nagy et al., 2006). Future research adopting the design of the current study might want to add additional measures to touch upon more sophisticated levels of morphological awareness, and see whether any significant contribution of morphological awareness to reading comprehension would emerge over and above vocabulary knowledge among adult English L2 learners. Methodologically, adding more measures or indicators of morphological awareness could help touch upon its latent structure and reduce measurement errors, even though SEM analysis can model relationships between variables with only one indicator. The different findings might also be attributed to variations in learners’ backgrounds, such as age and learning stage or L2 proficiency level, in the current study and previous studies. Studies that have found the unique contribution of morphological awareness to reading comprehension mostly focused on young children (Kieffer & Lesaux, 2008; Ku & Anderson, 2003; Wang et al., 2006, 2009) or adolescents (Nagy et al., 2006) that still demonstrated notable individual differences in decoding fluency. Nagy et al. (2006) found that morphological awareness contributed significantly to 8th and 9th grade English monolinguals’ rate of decoding morphologically complex words, after controlling for their decoding ability. Considering decoding efficiency could facilitate comprehension, the unique effect of morphological awareness on reading comprehension over and above vocabulary knowledge, as found in Ku and Anderson (2003) and Nagy et al. (2006) on English monolinguals and Kieffer and Lesaux (2008) on young ESL learners, could reflect the shared variance of decoding efficiency with morphological awareness and reading comprehension (Nagy, 2007). In the current study and Qian (1999), however, participants, albeit different in their context of English learning (EFL in China vs. ESL in Canada), were both adult learners with many years of formal English learning and relatively high English proficiency. Therefore, there was possibly limited individual difference in those learners’ English decoding fluency. This might help explain why no significant proportion of the variance in reading comprehension was uniquely explained by morphological awareness/knowledge in the current study as it 123 1214 D. Zhang, K. Koda has been observed in previous studies with young children and adolescents. However, neither the present study nor Qian (1999) adopted decoding fluency as an outcome measure in modeling the relationship between morphological awareness/knowledge and reading comprehension. Future research might consider adding decoding fluency as another variable and enrich the model tested in the present study. As the first of its kind that modeled the intricate relationships of morphological awareness to lexical inferencing ability, vocabulary knowledge, and reading comprehension among adult EFL learners, our study appears to have a few limitations that warrant attention in future research. To begin with, although our study involved a significant number of participants, a larger sample would be desirable for future studies. In this study, data of the 113 participants for SEM analysis showed multivariate normality; fit indices (i.e., CFI and RMSEA) that are less sensitive to model complexity and sample size were used; data-based resampling was conducted. All of these have in one way or another helped us overcome, or at least minimize the impact of small sample size and get generally unbiased estimates of parameters. However, a larger sample is still preferred for analysis within the SEM framework, though there has not been any consensus in the literature about an exact sample size for conducting SEM analysis. There is another issue not dealt with in the current study but might be of interest for future research: whether the predictive relations observed in the current study with advanced adult Chinese EFL readers would hold for other groups of English learners, given variations in the nature of lexical exposure among learners at different stages of English learning and the level of linguistic distance between L1 and English among learners of different L1 backgrounds? Empirically, to answer this question, statistical testing by use of multiple-group SEM analysis would be necessitated. Future research might want to simultaneously consider different learner groups—same L1 but different L2 proficiency, and/or similar L2 proficiency but different L1—and directly compare the relational patterns of the groups to see whether the patterns would be invariant or different across the learner groups. References Anglin, J. M. (1993). Vocabulary development: A morphological analysis. Monographs of the Society for Research in Child Development, 58(10), 1–165. Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173–1182. Bentler, P. M. (2005). EQS 6 structural equations program manual. Encino, CA: Multivariate Software, Inc. Carlisle, J. F. (2000). Awareness of the structure and meaning of morphologically complex words: Impact on reading. Reading and Writing: An Interdisciplinary Journal, 12, 169–190. Carlisle, J. F. (2003). Morphology matters in learning to read: A commentary. Reading Psychology, 24, 291–322. Cheung, G. W., & Lau, R. S. (2008). Testing mediation and suppression effects of latent variables: Bootstrapping with structural equation models. Organization Research Methods, 11, 296–325. Deacon, S. H., & Kirby, J. B. (2004). Morphological awareness: Just “more phonological”? The roles of morphological and phonological awareness in reading development. Applied Psycholinguistics, 25, 223–238. 123 Contribution of morphological awareness and lexical inferencing ability 1215 Gu, Y., & Johnson, R. K. (1996). Vocabulary learning strategies and language learning outcomes. Language Learning, 46, 643–679. Haastrup, K. (1991). Lexical inferencing procedures or talking about words: Receptive procedures in foreign language learning with special reference to English. Tubingen: Gunter Narr Verlag. He, L. (2010). The Graduate School Entrance English Examination. In L. Cheng & A. Curtis (Eds.), English language assessment and the Chinese learner (pp. 145–157). New York, NY: Routledge. Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1–55. Jin, Y., & Yang, H. Z. (2006). The English proficiency of college and university students in China: As reflected in the CET. Language, Culture and Curriculum, 19, 21–36. Kenny, D. A., Kashy, D. A., & Bolger, N. (1998). Data analysis in social psychology. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.), The handbook of social psychology (4th ed., Vol. 1, pp. 233–265). Boston, MA: McGraw-Hill. Kieffer, M. J., & Lesaux, N. K. (2008). The role of derivational morphology in the reading comprehension of Spanish-speaking English language learners. Reading and Writing: An Interdisciplinary Journal, 21, 783–804. Kline, R. B. (2005). Principles and practices of structural equation modeling (2nd ed.). New York: The Guilford Press. Ku, Y.-M., & Anderson, R. C. (2003). Development of morphological awareness in Chinese and English. Reading and Writing: An Interdisciplinary Journal, 16, 399–422. Kucera, H., & Francis, W. N. (1967). Computational analysis of present-day American English. Providence, RI: Brown University Press. Kuo, L.-J., & Anderson, R. C. (2006). Morphological awareness and learning to read: A cross-language perspective. Educational Psychologist, 41, 161–180. Laufer, B. (1992). How much lexis is necessary for reading comprehension? In P. J. L. Arnaud & H. Bejoint (Eds.), Vocabulary and applied linguistics (pp. 126–132). Basingstoke: Macmillan. McBride-Chang, C., Tardif, T., Cho, J.-R., Shu, H., Fletcher, P., Stokers, S. F., et al. (2008). What’s in a word? Morphological awareness and vocabulary knowledge in three languages. Applied Psycholinguistics, 29, 437–462. Mochizuki, M., & Aizawa, K. (2000). An affix acquisition order for EFL learners: An exploratory study. System, 28, 291–304. Nagy, W. (2005). Why vocabulary instruction needs to be long-term and comprehensive. In E. H. Hiebert & M. L. Kamil (Eds.), Teaching and learning vocabulary: Bring research to practice (pp. 27–44). Mahwah, NJ: Lawrence Erlbaum Associates. Nagy, W. (2007). Metalinguistic awareness and the vocabulary-comprehension connection. In R. K. Wagner, A. E. Muse, & K. R. Tannenbaum (Eds.), Vocabulary acquisition: Implications for reading comprehension (pp. 52–77). New York, NY: The Guilford Press. Nagy, W. E., & Anderson, R. C. (1984). How many words are there in printed school English? Reading Research Quarterly, 19, 304–330. Nagy, W. E., Berninger, V., & Abbott, R. (2006). Contributions of morphology beyond phonology to literacy outcomes of upper elementary and middle-school students. Journal of Educational Psychology, 98, 134–147. Nation, I. S. P. (1990). Teaching and learning vocabulary. New York, NY: Newbury House. Paribakht, T. S., & Wesche, M. (1999). Reading and “incidental” L2 vocabulary acquisition: An introspective study of lexical inferencing. Studies in Second Language Acquisition, 21, 195–224. Qian, D. D. (1999). Assessing the roles of depth and breadth of vocabulary knowledge in reading comprehension. Canadian Modern Language Review, 56, 282–307. Qian, D. D. (2004). Second language lexical inferencing: Preferences, perceptions and practices. In B. Laufer & P. Bogaards (Eds.), Vocabulary in a second language: Selection, acquisition and testing (pp. 155–169). Amsterdam: John Benjamins. Read, J. (1993). The development of a new measure of L2 vocabulary knowledge. Language Testing, 10, 355–371. Read, J. (1998). Validating a test to measure depth of vocabulary knowledge. In A. J. Kunnan (Ed.), Validation in language assessment (pp. 41–60). Mahwah, NJ: Lawrence Erlbaum. Sandra, D. (1994). The morphology of the mental lexicon: Internal word structure viewed from a psycholinguistic perspective. Language and Cognitive Processes, 9, 227–269. Schmitt, N., & Meara, P. (1997). Researching vocabulary through a word knowledge framework: Word associations and verbal suffixes. Studies in Second Language Acquisition, 19, 17–36. 123 1216 D. Zhang, K. Koda Schmitt, N., Schmitt, D., & Clapham, C. (2001). Developing and exploring the behavior of two new versions of the Vocabulary Levels Test. Language Testing, 18, 55–88. Shrout, P. E., & Bolger, N. (2002). Mediation in experimental and nonexperimental studies: New procedures and recommendations. Psychological Methods, 7, 422–445. Sobel, M. E. (1982). Asymptotic confidence intervals for indirect effects in structural equation models. Sociological Methodology, 13, 290–312. TEM-8 Development Committee. (2003). A collection of TEM-8 (Test for English Majors) test papers. China: Shanghai Foreign Language Education Press. Wang, M., Cheng, C., & Chen, S.-W. (2006). Contribution of morphological awareness to ChineseEnglish biliteracy acquisition. Journal of Educational Psychology, 98, 542–553. Wang, M., Ko, I. Y., & Choi, J. (2009). The importance of morphological awareness in Korean-English biliteracy acquisition. Contemporary Educational Psychology, 34, 132–142. Wu, A. D., & Zumbo, B. D. (2008). Understanding and using mediators and moderators. Social Indicators Research, 87, 367–392. Wysocki, K., & Jenkins, J. R. (1987). Deriving word meanings through morphological generalization. Reading Research Quarterly, 22, 66–81. 123