4.2 Preliminary results
4.2.2 Bivariate correlations
reading development in the first- and second-time data exhibited a consistent relationship.
The correlation results on cognitive component and reading ability agree with the correlation structure in Messick’s (1989) study, which involved an examination of
intercorrelations to determine how strongly each item correlates with the other items from its respective subscale. The cut-off criterion used by the author was a correlation of at least 0.20 between the item and most of the other items from its respective subscale. In the current study, cognitive components were measured on the basis of morphological production, morphological structure, rapid colour naming, rapid number naming, rapid word
segmentation and phoneme isolation, over both the datasets, were correlated significantly with reading comprehension and word recognition (range, 0.286 to 0.964). Rapid colour naming and rapid number naming were negatively correlated with morphological production, morphological structure, rapid word segmentation, phoneme isolation, reading
comprehension and word recognition at times 1 and 2 (Table 8), indicating that if students are highly competent in certain skills (morphological production, morphological structure, rapid word segmentation, phoneme isolation, reading comprehension and word recognition), then they require less time to name objects (colour and number in this study).
56 Table 8
Bivariate correlations amongst the cognitive and reading ability tests.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
1. RECT1 1
2. WORT1 .750** 1
3. RNCT1 –.792** –.634** 1
4. RNNT1 –.735** –.600** .905** 1
5. RWST1 .746** .637** –.778** –.797** 1
6. PHIT1 .745** .628** –.756** –.751** .886** 1
7. MAPT1 .747** .588** –.742** –.742** .783** .739** 1
8. MAST1 .732** .579** –.742** –.729** .770** .730** .964** 1
9. RECT2 .853** .678** –.895** –.874** .796** .765** .749** .752** 1
10. WORT2 .767** .596** –.790** –.750** .698** .662** .646** .656** .884** 1
11. RCNT2 –.406** –.295** .454** .417** –.369** –.336** –.320** –.317** –.456** –.445** 1
12. RNNT2 –.404** –.285** .401** .410** –.364** –.340** –.297** –.286** –.458** –.434** .668** 1
13. RWST2 .518** .377** –.532** –.495** .438** .406** .423** .409** .559** .539** –.729** –.717** 1
14. PHIT2 .463** .317** –.497** –.468** .408** .395** .367** .379** .518** .512** –.633** –.631** .714** 1
15. MAPT2 .491** .378** –.457** –.402** .415** .383** .350** .348** .469** .429** –.693** –.732** .701** .615** 1
16. MAST2 .422** .288** –.411** –.358** .353** .353** .298** .298** .428** .428** –.613** –.583** .667** .798** .674** 1
57
** p <0.01
Note. 1 = reading comprehension (REC), 2= word recognition (WOR), 3= rapid colour naming (RCN), 4 = rapid number naming (RNN), 5 = rapid word segmentation (RWS), 6 = phoneme isolation (PHI), 7= morphological production (MAP), 8= morphological structure (MAS). T1 = data from time 1, T2 = data from time 2.
58 4.3 Structural model: Cross-lagged modelling
The cross-lagged structural equation model (SEM) was analysing to answer the research question. The idea behind cross-lagged modelling is to identify the causal priority of variables by comparing the crossed and lagged paths between constructs. The crossed path coefficient represents the longitudinal prediction of one construct by another, above and beyond the autoregressive prediction of that construct from an earlier measure of itself (Curran & Bollen, 2001). For model estimation, SEM with the time 1 and time 2 data was conducted. The hypothesised model is shown in Figure 5. This model yielded an acceptable fit with the data (χ2/df = 346.142/92, GFI = 0.894, CFI = .962, RMSEA = .088). Figure 5 shows the complete model used for hypothesis testing.
Figure 5. Model for hypothesis testing.
Following the examination of the overall model fit indices, individual parameter estimates were assessed to analyse the feasibility of their estimated values. No individual parameters with unreasonable estimates ‘falling outside of acceptable ranges’ (i.e. standard errors that are extremely large or small, correlations greater than 1 and negative variance)
59
(Byrne, 1994) were found. For significance tests, parameter estimates were divided by standard error to determine whether the parameter estimates of the SEM statistically differ from zero. As indicated by an alpha level (p) of the .05 two-tail standard, if test statistics are greater than ±1.96, data are in the critical region and that the value of sample statistics differs significantly from zero.
4.4 Hypothesis testing
The aim of this study was not to build the best model of cognitive component and reading ability but to examine the hypothesized relationships between cognitive component and reading ability. Because the model fit is satisfactory, its regression coefficients were examined to validate hypotheses 1–6.
Table 9
Standardized path coefficients of the relationships.
Estimate
SE CR P
Standardized Unstandardized
COG T2 <--- COG T1 0.332 0.221 0.083 2.672 **
COG T2 <--- Read T1 –0.274 –0.032 0.014 –2.218 *
Read T2 <--- Read T1 0.201 0.208 0.062 3.385 ***
Read T2 <--- COG T1 –0.768 –4.581 0.371 –12.363 ***
COG T1 <--> Read T1 –0.883 –0.002 0.000 –10.353 ***
COG T2 <--> Read T2 –0.157 0.000 0.000 –2.226 * Note 1. REC = reading comprehension, WOR = word recognition, COG = cognitive component, READ = reading ability, T1 = data from time 1 and T2 = data from time 2.
Note 2. *** = p< .001, ** = p< .01, * = p< .05
60
Table 9 reports the standardised path coefficients of the relationships amongst the major variables without covariates in three types of the relationship including;
(1) Auto-regressive effects of the two variables.
The auto-regressive effect (C1C2) of the cognitive component at time-1 (COG T1) on the cognitive component at time-2 (COG T2) was significantly correlation, with the standard coefficient 0.332 (p< .01). And the auto-regressive effect (R1R2) of the reading ability at time-1 (REC T1) on the reading ability at time-2 (REC T2) was significantly correlation, with the standard coefficient 0.202 (p< .001).
(2) Synchronous correlations between the two variables.
The unconditional zero order correlation (C1R1) between the cognitive component (CCG T1) and the reading ability (REC T1), both at time-1 was significantly negative correlation, with the standard coefficient -0.883 (p< .001). And the zero order correlation (C2R2) between the cognitive component (COG T2) and the reading ability (REC T2), both at time-2 was significantly negative correlation, with the standard coefficient -0.157 (p< .05).
These negative correlations indicate an inverse relationship between the cognitive component time-1 and reading ability time-1, and also inverse relationship between the cognitive time-2 and reading ability time-2.
(3) Reciprocal effects between the two variables over time.
The Reciprocal effects (C1R2) of the cognitive component at time-1 (COG T1) on reading ability at time-2 (REC T2) was significantly negative relationship, with the standard coefficient -0.768 (p< .001). And the effect (R1C2) of reading ability at time-1 (REC T1) on cognitive component at time-2 (COG T2) was significantly negative relationship, with the standard coefficient -0.274 (p< .05). These negative effected indicate an inverse relationship between the cognitive component time-1 and reading ability time-2, and also inverse
relationship between the reading ability time-1.and cognitive time-2
61
Figure 6 illustrates the standardised path coefficients of the relationship between
cognitive component and reading ability for hypothesis testing by cross-lagged SEM analysis.
Figure 6. Standardised path coefficients of the relationships determined for hypothesis testing by cross-lagged SEM.
The negative correlations of C1R1, C2R2, R1C2, and C1R2 may involve with the negative correlation between rapid naming tests (rapid colour naming and rapid number naming) and the other tests. In order to assess the causal relationship from three cognitive components (morphological awareness, decoding-related and rapid naming), a cross-lagged SEM was analyses again by separated the cognitive variables as a latent factors, and assess the relationship between three cognitive variables and reading ability.
COG T1
Read T2 COG T2
C1C2 =0.332**
C1R2 = –0.768***
C2R2 = –0.157*
C1R1 = –0.883***
R1R2 = 0.201***
R1C2 = –0.274*
Read T1
62
Figure 7. The model of the relationships between morphological awareness, decoding-related and rapid naming, and reading ability.
Table 10
Standardized path coefficients of the relationships.
Estimate
63
Table 10 (continued). Standardized path coefficients of the relationships.
Estimate
SE CR P
Standardized Unstandardized
MA2 <--- Read1 0.093 0.003 0.083 0.035 0.075
Read2 <--- MA1 0.185 0.097 0.064 1.505 *
Read2 <--- DE1 0.154 0.606 0.619 0.980 **
Read2 <--- RN1 -0.267 -2.084 0.646 -3.228 **
DE2 <--- MA1 -0.023 -0.024 0.068 -0.358 0.721
RN2 <--- MA1 0.265 0.042 0.010 3.978 ***
MA2 <--- DE1 0.197 1.002 0.675 1.484 *
RN2 <--- DE1 -0.055 -0.086 0.099 -0.864 0.387
MA2 <--- RN1 -0.198 -0.699 0.692 -1.009 *
DE2 <--- RN1 -0.158 -1.131 0.671 -1.685 *
MA1 <--> DE1 0.455 0.001 0.000 7.602 ***
MA1 <--> RN1 -0.580 -0.001 0.000 -9.102 ***
DE1 <--> RN1 -0.524 0.000 0.000 -8.321 ***
DE1 <--> Read1 0.469 0.001 0.000 7.156 ***
RN1 <--> Read1 -0.769 -0.002 0.000 -9.456 ***
MA1 <--> Read1 0.490 0.007 0.001 7.458 ***
RN2 <--> Read2 -0.097 0.000 0.000 -1.567 0.117
DE2 <--> Read2 -0.021 0.000 0.001 -0.355 0.722
DE2 <--> Read1 0.132 0.000 0.001 0.539 *
DE2 <--> RN2 -0.587 -0.001 0.000 -7.372 ***
MA2 <--> RN2 -0.556 -0.001 0.000 -6.956 ***
MA2 <--> DE2 0.775 0.009 0.001 9.400 ***
64 Note. *** = p< .001, ** = p< .01, * = p< .05
The figure 7 shows that the model of the relationships between morphological awareness, decoding-related and rapid naming, and reading ability without covariates. The table 10 results show that the rapid naming was significantly negative relationship with other variable (e.g. rapid naming at time 1 has negative relationship with reading ability at time 2 with the standardised path coefficients is -0.267). These result indicated that if student have high ability in morphological awareness decoding-related variables and reading ability then they require less time to do rapid naming test.
However, when combine rapid naming, morphological awareness and decoding-related variables into the cognitive component, it can be assumed that cognitive component was significantly related with reading ability over the time and also reading ability was
significantly related with cognitive component over the time. In sum, the Table 11 reports the summary of finding of the cross-lagged SEM results with the hypothesis of this study.
Table 11
Summary of findings.
Hypotheses Findings
1. Cognitive component and reading ability are correlated with each other in time 1.
1. Supported: Cognitive component was significantly related to reading ability in time 1 (C1R1 =–0.883, p< .001).
2. Cognitive component and reading ability are correlated with each other in time 2.
2. Supported. Cognitive component was significantly related to reading ability ( C2R2 = –0.157, p< .05).
65 Table 11 (continued)
Summary of findings.
Hypotheses Findings
3. Cognitive component in time 1
predicted cognitive component in time 2.
3. Supported: Cognitive component at time 1 predicted cognitive component at time 2 (C1C2 = 0.332, p< .01).
4. Reading ability in time 1 positively predicts reading ability in time 2.
4. Supported: Reading ability at time 1 predicted reading ability at time 2 (R1R2 = 0.201, p< .001).
5. Reading ability in time 1 predicts cognitive component in time 2.
5. Supported: Reading ability in time 1 significantly predicted cognitive component in time 2 (C1R2 = –0.274, p< .05).
6. Cognitive component in time 1 predicts reading ability in time 2.
6. Supported: Cognitive component in time 1 significantly predicted reading ability in time 2 ( C2R2 = –768, p< .001).
66
CHAPTER FIVE
DISCUSSION
5.1 Summary
The relationship between cognitive component and reading ability has been studied from several perspectives but remains inadequately understood, especially in the context of the Thai language, which is characterised as a transparent orthography. Several models have been employed in explaining the reading process, in which the relationships amongst reading-related variables are analyzed in a complex manner, for example the simple view of reading (Gough & Tunmer, 1986), Bottom-up Approach Model (Carrell, 1988), Top-down Approach Model (Ausubel, 1956) and Interactive View of reading (Rumelhart (1990). This study sought to explore these relationships in the context of Thai—a direction that has not been previously pursued. This study looked into the relationship between cognitive component and reading ability in reading development to determine whether specific directionality patterns exist over a period of one calendar year in the learning process.
Two tests were developed to assess each of the three domains of cognitive component in the context of the Thai language. That are, the morphological production and
morphological structure tests were developed for the morphological awareness domain; the rapid colour naming and rapid number naming tests were created for the rapid automatized naming domain; and the phoneme isolation and rapid word segmentation tests were designed for the decoding-related domain. Two other tests were developed to assess reading ability: the reading comprehension and word recognition tests. The content validity of all the tests was evaluated by an expert panel. The reliability of all the tests was greater than 0.60 (ranged from 0.647 to 0.981).
67
The cognitive component and reading ability of the 357 students in the fourth grade were examined to determine how cognitive and reading ability relationships change over time when these students reached grade five. As previously indicated, first-time data collection was conducted in January–February 2014 (time 1), and second-time data collection was carried out in November 2014–January 2015 (time 2). Two-time cross-lagged panel models were assessed to ascertain the directionality of reading development.
After data collection, the frequency distribution of cognitive component and reading ability across all the eight tests for time-1 and time-2 data was evaluated in term of skewness and kurtosis statistics, which suggest that all the tests were characterised by normal
distribution. The intercorrelations between cognitive component and reading ability in times 1 and 2 two were significant (ranged from 0.286 to 0.964). In order to achieve the aim of the study, the cross-lagged SEM was used to determine the direction of the relationship between cognitive component and reading ability in time-1 and time-2 data. The following sections discuss the major findings and related considerations.
5.2 Major findings and discussion
5.2.1 Synchronous correlations between cognitive and reading ability in time 1 and time 2 (hypotheses 1 and 2). As expected, hypotheses 1 and 2 are supported by the results;
cognitive component and reading ability were correlated at time 1 as well as at time 2, as indicated by the cross-lagged SEM analysis. These results are expected given previous studies undertaken in other languages (e.g. research in the Greek language by Kendeou, Papadopoulos and Kotzapoulou (2012). The findings on the bivariate correlations amongst the cognitive and reading ability tests show that all the cognitive tests (i.e. rapid colour naming, rapid number naming, rapid word segmentation, morphological production and morphological structure) were significantly correlated with the reading ability tests (i.e.
68
reading comprehension and word recognition). This relationship suggests that children with high cognitive ability tend to have higher reading skills than children with low cognitive ability. As noted earlier, the relationship between cognitive component and reading ability is a well-established psychological principle. Hypotheses 1 and 2 in this study align with the Simple View of Reading (SVR), which illustrates the relationship between language comprehension and decoding skills (Gough & Tunmer, 1986). Gough and Tunmer (1986) explained the SVR model as follows: reading comprehension (R) is the product of the relationship between language comprehension (C) and decoding skills (D) (R = C × D). The SVR equation illustrates that with insufficient decoding skills or language comprehension (or both), reading comprehension produces a value of zero if some other skill in the equation also registers a value of zero. Kendeou, Papadopoulos and Kotzapoulou (2013) empirically tested the emergence of SVR in the transparent orthography of the Greek language and found correlations between decoding skills (such as phonological awareness, letter identification, and vocabulary knowledge), and language comprehension skills (such as listening
comprehension). These patterns are broadly consistent with the SVR perspective that
decoding-related and listening comprehension skills in Greek are, at least to a certain extent, separable in young children. The relationship between cognitive component and reading ability is found not only in languages characterised by transparent orthographies but also in languages with opaque orthographies. Georgiou, Torppa, Manolitsis, Lyytinen and Parrila (2012) examined the longitudinal predictors of non-word decoding, reading fluency and spelling in three languages that vary in orthographic depth: Finnish, Greek and English. The authors found a significant correlation amongst phonological awareness, letter knowledge and rapid automatized naming speed in the three languages. The results of the
aforementioned studies and those of the current research both indicate that the relationship
69
between cognitive component and reading ability is important to reading development in all languages.
5.2.2 Auto-regressive effects between cognitive component and reading ability in times 1 and 2 (hypotheses 3 and 4). A notable stability in cognitive component and reading ability over time was found in this study. The cognitive component at time 1 was positively related to that at time 2 (hypothesis 3), and the reading ability at times 1 was positively related to the reading ability at time 2 (hypothesis 4). These results suggest that both
cognitive ability and reading development remained largely stable over the interval spent by students in acquiring reading eligibility for time 2. This finding is consistent with that of Winskel and Iemwanthong (2010), who investigated reading and spelling development in Thai children. The authors found that the children performed substantially better on word reading and spelling tests than on the corresponding non-word reading and spelling activities.
After 4 months of schooling, the grade 1, 2 and 3 children achieved 42% accuracy in word reading, 24% accuracy in reading the corresponding non-words, 32% accuracy in word spelling and 17% accuracy in non-word spelling. A noticeable result is that reading and spelling performance rapidly increased between the youngest grade 1 children and the grades 2 and 3 children. Nevertheless, these results are more comparable with those on children learning to read languages with relatively opaque orthographies, such as English (Stuart &
Coltheart, 1988; Wimmer & Goswami, 1994), rather than languages with more transparent orthographies, such as German (Goswami, Ziegler, Dalton & Schneider, 2003; Landerl, Wimmer, & Frith, 1997). Children learning to read German rapidly develop the ability to read non-words because they have ready access to grapheme–phoneme conversion rules, which are compatible with the requirements of non-word reading tests in German (Landerl et al., 1997). By contrast, English children slowly develop such ability in early age because of the inconsistency of English orthography and the over-reliance on the use of large grain sizes,
70
the adoption of whole-word or lexical strategy and the presence of ‘a rather underdeveloped alphabetic strategy’ for reading non-words (Wimmer & Goswami, 1994). The results of the current study reflect that previous cognitive component/reading abilities may predict such abilities over time.
5.2.3 Cross-lagged relationships between cognitive component and reading ability over the time (hypotheses 5 and 6). The findings of this work support the hypotheses;
significant relationships between cognitive and reading ability were found by the SEM cross-lagged analysis. The reading ability at time 1 was significantly associated with the cognitive component in time 2 (hypothesis 5), and the cognitive component in time 1 predicted the reading ability at time 2 (hypothesis 6). These results are aligning with the theories on the top–down processing and traditional bottom–up approach to reading. Moreover, the direction of relationship between cognitive component and reading was significantly and reciprocal in nature; that is, cognitive component could predict reading ability and vice versa over time.
This suggests that reading development in the Thai language benefits not only from
traditional bottom–up and top–down processing but also from an interactive view of reading.
Thai is a complex language. For example, its writing structure differs from its spelling structure, and a complex vowel can be placed above, below or on either side of a consonant.
Sounds can also be combined to produce a large additional number of vowels. The traditional bottom–up approach to reading can explain how Thai children learn to read. Thai children have to learn the phonemes of Thai consonants, vowels and tone, after which they are required to learn how to combine consonants and vowels to pronounce a word. This process indicates that Thai children develop cognitive component beginning from small units and then apply this skill to reading and comprehension. Figure 8 illustrates the process by which consonants, vowels and tone markers are combined to enable reading of a Thai word.
71
เรื่อง
(Raung) /Story
ร เ-ื อ ง ื
Initial
consonant Vowel Final
consonant Tone marker
เรื่อง (Raung) /Story
Figure 8. Process for reading a Thai word.
Another peculiarity of Thai is that spaces are not used to segment syntactic units but only to delimit sentences; the language also rarely uses punctuation. When reading Thai, therefore, learners are required to segment using cues other than spaces. The lack of word boundaries makes word segmentation skills essential for Thai readers to achieve reading comprehension (Winskel, Perea, & Ratitamkul, 2012).
These findings are consistent with those of Chapman and Tunmer (2003), who support the traditional bottom–up approach to reading. The authors indicated that the development of successful reading skills necessitates that children employ efficient word recognition
strategies, which are necessary for the development of rapid word decoding skills. High levels of automaticity in word recognition, in turn, make available greater cognitive resources for allocation to comprehension and text integration processes (Adams, 1990; Tunmer &
Chapman, 1998). Moreover, finding of current study is consistent with the previous study of reading model in Chinese language (Yeung, Ho, Chan, Chung, & Wong, 2013). Yeung et al.
(2013)indicated that rapid automatized naming and morphological awareness predicted reading comprehension significantly by using word reading as a mediator factor among Chinese grade 4 students in Hong Kong. Yeung et al. (2013) suggested that morphological awareness, decoding-related skills and RAN were important cognitive factors in the reading
72
model of non-alphabetic languages. The same was also found in the present study with a transparent orthography, Thai.
Both traditional bottom–up and top–down processing can be applied in examining Thai reading. In this work, reading ability was significantly associated with cognitive component over time, indicating that top–down processing theory can explain reading development in the context of the Thai language. This type of processing is principally based on the prior knowledge of children and in the communicative tasks that they accomplish in everyday life.
To comprehend a message, children begin with meaning at the paragraph level, after which they proceed to understanding the sentences and words that constitute the paragraph. Top–
down processing therefore enables the understanding of an ambiguous text because it activates high-level schemas that guide the reading process. Under this backdrop, prior knowledge and reader expectations become essential elements in the comprehension process.
Thus, when children confront a text, their previous experience guides their comprehension process. Children can also use context clues to determine the meaning of words that function in more than one way. For instance, the word ดวง (‘duong’/‘horoscope’) denotes different
Thus, when children confront a text, their previous experience guides their comprehension process. Children can also use context clues to determine the meaning of words that function in more than one way. For instance, the word ดวง (‘duong’/‘horoscope’) denotes different