CHAPTER FIVE TEACHERS’ CONCEPTIONS OF EXCELLENT TEACHING AND TEACHING PRACTICE IN CHINESE MIDDLE SCHOOLS
Study Two provided the validity evidence for the survey instruments and for the measurement models of teachers‘ conceptions of excellent teaching and teaching practice. The statistical evidence for the two measurement models, however, was generated through the surveys with quite small sample sizes. Study Three confirmed these two measurement models with a large sample size and explored the structural relations between the two models.
Study Three reports on a large quantitative survey in which two acceptably-fitting TCET and TCTP models were identified. Based on these two models, an integrated structural model of teachers‘ conceptions of excellent teaching and teaching practice was developed. This study answers the two research questions that were posed at the end of the reviewed literature. First, the TCET model and the TCTP model were identified to answer the research question of how Chinese middle school teachers perceived excellent teaching and the teaching practices they used. The relationships between teacher characteristics and their conceptions were also reported. Second, the structural model of teachers‘ conceptions of excellent teaching and teaching practice was established. This model answered the research question regarding how Chinese middle school teachers‘ conceptions of excellent teaching were related to the practices they used.
The TCET and TCTP questionnaire used for this study was developed from the findings of Study Two and the reviewed literature. The statistical techniques, EFA and CFA, were used to analyse the quantitative data. The structural equation modelling (SEM) approach was used to test the relationships between TCET and TCTP within a model.
Sample size is critical in the SEM procedure. Velicer and Fava (1998) reported that sample size accounted for 23% of the variance in a summary index describing the discrepancy between an obtained and expected factor matrix. Anderson and Gerbing (1984) found that sample size had an impact on several fit indices such as Goodness-ofFit Index (Jöreskog & Sörbom, 1986), the Adjusted Goodness-of-Fit Index (Jöreskog & Sörbom, 1986), Root Mean Square Residual (Jöreskog & Sörbom, 1986), and the Nonnormed Fit Index (Bentler & Bonett, 1980; Tucker & Lewis, 1973). Costello and Osborne (2005) identified that sample size significantly affected the accurate solution rate and item misclassification. Gerbing and Anderson (1985) investigated the effects of several variables and found that sample size had the largest effect on the variance in parameter estimates. Bentler and Yuan (1999) argued that a large enough sample size was one of prerequisites to obtain meaningful parameter estimates when estimation procedures were used to calculate the model parameters. However, it does not seem credible that only sample size is considered. Some researchers have demonstrated that other points may affect sample size. For example, the ratio of variables to factors (Browne, 1968; MacCallum, Widaman, Preacher, & Hong, 2001; MacCallum et al., 1999; Preacher & MacCallum, 2002; Tucker, Koopman, & Linn, 1969), the level of communality (MacCallum et al., 1999; Tucker et al., 1969; Velicer et al., 1982). One of most popular recommendations is that the ratio (N:p) of number of subjects (N) to number of variables (p) is a better way to determine a minimum sample size (Kline, 1994). Further, Osborne and Costello (2004) reported that the N:p ratio was a consistent predictor of stability in factor structures, the occurrence of Type I errors, and the ‗correctness‘ of factor structures. They also reported a relative lack of unique impact of the absolute N after the N:p ratio was accounted for. Therefore, the N:p ratio was considered for the analysis of this study rather than simply the absolute sample size alone.
Some researchers have recommended various N:p ratios which ranged from 2:1 through 20:1 for factor analysis. Cattell (1978) recommended 6:1, Everitt (1975) and Garson (2008) 10:1, and Hair, Anderson, Tatham, and Black (1995) 2:1 to 20:1. Ford, MacCallum, and Tait (1986) examined the articles published in the journals of Applied Psychology, Personnel Psychology, and Organizational Behaviour and Human Performance during the period of 1974-1984. They found that 56% of the N:p ratios in reviewed articles were less than 10:1. Osborne and Costello (2004) summarised PsychINFO articles within two years that reported 63% of the N:p ratios as being less than 10:1 and 79% of the ratios as being less than 20:1. Costello and Osborne (2005) examined how N:p ratios affected the accuracy of solutions (e.g., producing correct factor structure and the number of items misclassified) by creating the ratios of 2:1, 5:1, 10:1, and 20:1 in twenty samples. They found that the sample with the larger N:p ratios such as 10:1 and 20:1 tended to produce more accurate solutions and results. In order to obtain accurate procedures and robust results, the sample size in this study was 951, the ratio of number of participants to number of variables was about 16:1 (951/58) for TCET and 19:1 (951/48) for TCTP.
A quantitative instrument, that is teachers‘ conceptions of excellent teaching and teaching practice in Chinese middle schools, was used in this study. This 106-item questionnaire was developed based on Study Two and the reviewed literature. The questionnaire included three parts. Part One was about the statements of excellent teaching in Chinese middle schools. Part Two was about the statements of teaching practice in Chinese middle schools. The explanations for the requirements and response scales were the same as those in Study Two. In Part Three, teachers were asked to give some personal information such as sex, age, teacher certificate, qualification, and the subjects and grades they taught. Completion of the questionnaire was expected to take about 35 minutes (18 minutes for Part One, 15 minutes for Part Two, and two minutes for demographic questions (see Appendix G for English and Chinese versions). Procedures for ensuring the confidentiality and anonymity of participants were similar to those in Study One.
The questionnaire was originally designed in English. The researcher translated the English version into Chinese. The three ITS guidelines explained in Chapter Three were used in this process. The Chinese colleague mentioned in Study One and Study Two back-translated the questionnaire. The two English versions were compared. All sentences had equivalent meanings in these two versions. For the final Chinese version, a colleague also helped the researcher check Chinese language adaptations such as words, idioms and phrases more suited to local language usage in the north of China (this judge was from northern China). Any disagreements with the Chinese translation were discussed until overall agreements were achieved in both language and logical meaning.
A convenience sample of 2200 Chinese middle school teachers, from 29 middle schools in nine cities of Liaoning Province in the north of China, was approached in August of 2008. By the end of October, the researcher had received 1051 questionnaires back with a response rate of 48%. Among these respondents, 100 participants were dropped because they had too many missing values and/or high rate of agreements to answer questions (see details in data cleaning), and 951 questionnaires were kept for analysis. The majority of the teachers held Bachelor qualifications (81%) or were female (62%). Just over 40% of the teachers were aged between 33-40 years or held intermediate teacher certificates. Approximate 30% of the teachers had 8 or 8-15 years work experience. The equal percentage of the teachers taught Grade 7, Grade 8, or Grade 9 respectively. Only 2.5% taught across three grades (see Table 17). 5.4.2 Administration
The survey was conducted by three qualified research assistants, known to the researcher, in Liaoning Province in the north of China. Before conducting the survey, the researcher gave clear instructions to guide them through telephone and email. Since they had helped the researcher conduct the two previous surveys, they knew clearly what to do and how to do it.
The survey process was similar to those in Study One and Study Two. In August 2008, the Chinese research assistants personally contacted 29 middle schools in nine cities in Liaoning Province in the north of China. Research assistants visited each school that had agreed to cooperate on the telephone. They briefed each principal with a written Participant Information Sheet (PIS), and asked for permission to conduct the research within the school. Once the principal consented by signing a consent form, the principal was asked to call for teacher volunteers within the school. After that, the principal distributed the Teacher Participant Information Sheets, Questionnaires, and addressed/stamped return envelopes to teacher volunteers in the school. Teachers were asked to return them directly to the research assistants, or to drop them into a special box within four weeks using the addressed stamped envelopes supplied with the questionnaires. The research assistants collected the boxes from the schools. The ways of achieving the confidentiality and anonymity of the participants were similar to those used in Study One in Chapter Three.
This section reviews four procedures involved in the data analysis process. These are described under the subheadings of data inputting, data cleaning, data distribution checking, and data analysis.
18.104.22.168 Data Inputting
At the end of October 2007, the researcher assistants received 1051 questionnaires. A research assistant collected all questionnaires from the other two research assistants and was responsible for data input in China. She numbered all questionnaires and assigned identification numbers to each school. She also organised to input the data into a spreadsheet program in Excel 2007 by hiring ten university students. The research assistant divided these ten students into five groups. Four groups entered 200 questionnaires and one group entered 251 questionnaires. When the groups finished data inputting, they exchanged and checked the other groups‘ inputting. Once the data had been checked, the research assistant merged each group‘s data into one data set and sent it to the researcher through email in New Zealand. The researcher converted the data from Excel into SPSS 17.0. Note that when the participants chose two answers for a variable or did not respond, the value was defined as a missing value.
22.214.171.124 Data Cleaning
There were three steps to check the validity of data entry. First of all, when the data inputting was finished, the data entry accuracy was checked by one of the other groups. Any mistakes found were revised by referring back to the original questionnaires. In addition, the research assistant, who was responsible for data input, hand-checked a 10% sample of data (1051*10=105 questionnaires) with the original questionnaires. No incorrect entries were found. Finally, the valid ranges of values were established (i.e., maximum and minimum). Illegal values were corrected by checking the original questionnaires.
As with Study Two, missing values were checked for variables (items) and cases (teachers). Since 106 variables had less than 5% missing values, all items in the questionnaire were kept for further analysis.
The cases, however, with more than 5% missing values were dropped. Because of missing values, 100 questionnaires were dropped. The questionnaire of 106 items consisted of 58 TECT items and 48 TCTP items. Cases were dropped if they had more than 5% missing values: three for TCET or three for TCTP. In total, 66 cases were dropped because they had greater than 5% missing values for either section: 46 were dropped from the TCET, 26 cases from the TCTP, and six cases for both. Additionally, the cases which gave a 100% the same answer for every item were dropped on the assumption that identical responses indicated a lack of honest response. A total of 34 cases were dropped: 19 from the TCET, 20 from the TCTP, and five from both. Therefore, a total of 100 cases were discarded for too much missing or a lack of genuine response. Consequently, 951 cases were kept for analysis giving a N:p ratio of approximately 16:1 (951/58) for the TCET items and 20:1 (951/48) for the TCTP items. Note that TCET items and TCTP items were separated for analysis using the sample of 951.
Before checking the data distribution, the remaining cases with no more than three missing values were imputed using the expectation maximisation (EM) missing values procedure (Dempster, Laird, & Rubin, 1977). The EM procedure may legitimately estimate an extreme value that exceeds the actual minimum of 1.0 and maximum of 6.0 in the response scale for the TCET items and the actual maximum of 5.0 for the TCTP items. Therefore, the values that exceeded the actual minimum of 1.0 and maximum of 6.0 or 5.0 in the response scale were appropriately corrected to the minimum or maximum values.
126.96.36.199 Data Distribution Checking
The data description for the TCET and TCTP items was checked. Mean scores for the TCET items ranged from 3.60 to 5.47, standard deviations ranged from .84 to 1.63, skewness values ranged from .09 to -1.95, and kurtosis values ranged from .02 (ET42) to 4.24 (ET37). Mean scores for the TCTP items ranged from 3.03 to 4.41, standard deviations ranged from .77 to 1.39, skewness values ranged from -.08 to -.96, and kurtosis ranged from -.02 to -1.27. Thus, it was concluded that the variables had a sufficiently normal distribution for further analysis.
188.8.131.52 Data Analysis
The TCET and TCTP items in two measurement models were estimated independently. Exploratory factor analysis with maximum likelihood estimation and oblimin rotation was conducted to establish a probable measurement model for each inventory. As with Study Two, when using EFA, a factor was retained if it had at least 3 statements whose loadings were greater than .30. The statement was retained if it had a clear logical connection with other statements in the same factor and its loading was greater than .30. Confirmatory factor analysis (CFA) was employed to test how well the models fitted with data. Structural equation modelling (SEM) was then used to explore how the TCET responses related to the TCTP‘s.
Based on the theory of planned behaviour (Ajzen, 2005) and the related previous empirical findings regarding the relationship between teachers‘ conceptions of teaching and their practices in Chapter Two, it was hypothesised firstly that teachers‘ conceptions of excellent teaching related to their conceptions of teaching practice. The second hypothesis was that the certain types of conceptions of excellent teaching would relate to the corresponding types of conceptions of teaching practice. It was expected that ‗teacher-centred‘ conceptions of excellent teaching would shape ‗teacher-centred‘ conceptions of teaching practice, and ‗student-centred‘ conceptions of excellent teaching would predict ‗student-centred‘ conceptions of teaching practice.
184.108.40.206 TCET Model
There were three steps to deal with the TCET items using EFA. Firstly, the data were analysed using EFA (i.e., maximum likelihood estimation (MLE) with oblique rotation) allowing all factors with eigenvalues >1.00 to be identified by specifying four sets number of factors (10, 9, 8, and 7). The reason for selecting these sets was that eight factors were identified in Study Two. The resulting factor patterns, however, were not as logical and were very different from those identified in Study Two. Therefore, this analytic method did not work.
The second method was to use the findings of Study Two. The items in eight factors were tested separately using EFA and all loadings were greater than .30. The 58 TCET items behaved as expected when each factor was evaluated separately. However, when the eight-factor, 58-item TCET model was tested simultaneously in CFA, the fit indices for the model were not good and some factor inter-correlations were very high (i.e., approximately half of correlations were >.90). Thus, this model was discarded.
Thirdly, 58 TCET items were analysed using EFA (i.e., MLE with oblique rotation) by specifying a different number of factors (i.e., between 8 and 3) to ensure optimal interpretation of all factors. Factor reduction led to a five-factor, 34-item model by removing 24 items. After identifying the TCET model using the EFA procedure, CFA was used to test how well this model fitted the data. In CFA, the items that caused negative error variance by being overly correlated with each other or which had low loadings (<.30) on their intended factors were removed. The instrument ended with 26 items and the problematic items were dropped. By discarding eight items, a hierarchical five-factor model of TCET was generated (2=1154.19; df=294; 2/df=3.93; RMSEA=.055, 90% CI= .052-.059; SRMR= .055; CFI=.91; and gamma hat=.94) (see Figure 5). Note that the correlation and error terms were removed for simplicity in Figure 5.
This well-fitting, two-order model consisted of five conceptions which could be placed into two major dimensions. Teacher As Examination-Oriented was perceived as belonging to the examination-oriented dimension. This conception, ranked in the second order, least correlated with the other four conceptions. The other four conceptions could be defined as the teacher-student dimension. They included Teacher As Developing Lifelong Learners, Teacher As Student Focused, Teacher As Responsible For Engaging Students In Learning, and Teacher As Professional Learner. These four conceptions ranked in the first order were highly correlated to each other. The five conceptions cooperatively contributed to the teachers‘ conceptions of excellent teaching in Chinese middle schools. All remaining item loadings were greater than .30 and each factor included at least three items. Item loadings indicated that the items were related to each other as a separate factor in the model.
TABLE OF CONTENTS
LIST OF TABLES
LIST OF FIGURES
LIST OF APPENDICES
CHAPTER ONE INTRODUCTION
1.1 Problem Definition of the Research
1.2 Aims of the Research
1.3 Chinese Education Context
1.4 Structure of the Thesis
CHAPTER TWO LITERATURE REVIEW
2.1 The Theory of Planned Behaviour
2.2 Conception and Teacher Conception
2.3 Teachers‘ Conceptions of Teaching
2.4 Teaching Practice
2.5 Links between Teachers‘ Conceptions of Teaching and Teaching Practices
2.6 Teachers‘ Conceptions of Excellent Teaching
CHAPTER THREE CONCEPTIONS OF EXCELLENT TEACHING AND SELFREPORTED TEACHING PRACTICES IN CHINESE MIDDLE SCHOOLS
3.1 Middle School Participants‘ Conceptions
3.2 Narrative Inquiry
CHAPTER FOUR MEASURING TEACHERS’ CONCEPTIONS OF EXCELLENT TEACHING AND TEACHING PRACTICE IN CHINESE MIDDLE SCHOOLS
4.1 General Discussion about Validity Procedures
4.2 TCET Questionnaire
4.3 TCTP Questionnaire
CHAPTER FIVE TEACHERS’ CONCEPTIONS OF EXCELLENT TEACHING AND TEACHING PRACTICE IN CHINESE MIDDLE SCHOOLS
5.1 Sample Size
CHAPTER SIX CONCLUSION
6.4 Future Research
GET THE COMPLETE PROJECT
Teachers‘ Conceptions of Excellent Teaching in Chinese Middle Schools