CHAPTER 4 STUDY ONE: QUESTIONNAIRE SURVEY
4.4 DATA ANALYSIS
The statistical analysis software SPSS was employed to analyse quantitative data, as it is effective in doing a variety of both descriptive and inferential statistical analysis. The primary stage of data analysis mainly involved clearing and organising the data sets. Firstly, incomplete questionnaires in which participants did not answer a single question or only filled in their demographic information but did not respond to the remaining sections of the questionnaire were deleted. Secondly, missing data were excluded and extreme outliers were identified through inspecting univariate data and omited, as they might interfere with the normal analysis process and mislead the results. Thirdly, all the variables were renamed in the SPSS dataset, since the original data downloaded from the Qualtrics website were unorganised and named with default codes from the system. At this stage, several characteristics of teachers were coded. More specifically, Chinese teachers were coded as value “1” and British teachers were coded as value “2”; and female teachers were coded as value “0” and male teachers were coded as value “1”. This
background information was used as categories for conducting comparisons between different groups. Moreover, all the rates on 5-point Likert scales were automatically numbered as 1, 2, 3, 4, and 5 for items with response from fairly serious/low/strongly disagree to extremely serious/high/totally agree in the Qualtrics dataset. The benefits of doing the initial data exploration and organisation are that it makes the data analysis processes easier and clearer, and also provides the researcher an overview of the profile of respondents and makes him/her more familiar with the data (Pallant, 2013).
Both descriptive and inferential statistical analyses were conducted, in order to reveal and present research findings in the most effective way. To be more specific, descriptive statistics were applied to illustrate the frequency, percentage and central tendency of the quantitative data in the questionnaire. For example, means and standard deviations of participants’ demographic data were calculated. In addition, instructors’ most frequently experienced student behaviours and the proportion of instructors who report one particular misbehaviour as being the most troublesome were also examined by descriptive statistics. Charts and figures were made to show frequency and percentage. These visualised results can attract attention from readers quickly and make it easy for readers to understand the findings. The descriptive statistics also depicted the trends of responses from participants, for example teachers’ perspectives of students’ behaviour and their emotional experiences through line graphs, and made the comparison between the two groups more straight and clear.
Inferential statistics were used to measure differences and correlations between the two groups. According to D'Agostino, Belanger and D'Agostino (1990), before
conducting inferential statistical analyses, the normality of the data needs to be tested. If the data are normally distributed, parametric analysis should be applied; otherwise, the researcher should use non-parametric analyses to examine the data. However, as Ghasemi,
& Zahediasl (2012) argue, the parametric analysis still works well on non-normally distributed data, if the sample size is greater than 30. Therefore, in the present research, results were reported mainly from the parametric analyses as the sample size of each group in the current study was greater than 40. The non-parametric analysis, Mann
Whitney U test, was still run to compare teachers’ emotional reactions to see if there were any differences between results from these two analyses. Through this the reliability of the findings in this research can be enhanced. In addition, although there might be a chance to encounter Type I errors since no adjustment (e.g. Bonferoni adjustments) has been done to correct the error, it might be good for avoiding the Type II errors, in case the analysis did not identify any significant differences among the results. For
complementing this weakness of the data analysis, the effect sizes (partial eta squared) for each ANOVA result were reported along with the value (p) of significance.
For analysing the results of teachers’ perceptions of 17 student misbehaviours, factor analysis, one-way ANOVA and correlation test were employed. As Fabrigar, Wegener, MacCallum, & Strahan (1999) point out, factor analysis is good at revealing latent
relationships among complicated concepts. It helps the researcher unveil similar response patterns to a buried factor underlying participants’ responses (Costello & Osborne, 2005). Therefore, hidden categories of all 17 student behaviours may be discovered through the factor analysis based on teachers’ responses and it would make the results more explicit and meaningful for further discussions.
Multivariate analysis of variance (MANOVA) is employed to test the influence from a variety of independent variables (e.g. gender, and country of birth) on instructors’ difference emotions (DVs). Following it, a series of univariate ANOVAs were conducted to detect the specific differences on means of several groups. For instance, using ANOVA to compare 5 categorised groups of instructors’ years of teaching experience and then to
determine if instructors’ emotional experiences vary when it comes to their years of teaching in the UK. Furthermore, for examining relationships between participants’ characteristics and their perceptions and feelings, Pearson’s correlation test was applied. The reason for choosing it was that the causality relationship between variables (e.g. age and emotional experiences) in this research is not clear; more specifically, one variable as an X variable was not experimentally manipulated to see the effect from the Y variable, as it did not matter which variable was the X variable in the current research. As such, according to Field (2013), instead of using the linear regression that clearly defines which variable is X and aims to predict the Y from it, the correlation test is more appropriate to use if the researcher measures both variables (Field, 2013).
At last, there were also a number of qualitative results from open-ended questions appearing in the questionnaire results. They were downloaded and exported into word documents. Similar qualitative analysis methods (e.g. thematic analysis) to those used in analysing interview data were employed in this section as well. However, there might be discovered some categories rather than themes from the data, because answers in the open-ended questions were very short most of the time.