*3.1. Descriptive Statistical Analysis of Pre-Service Teacher Preparation Scale*

Since some teachers responded to the pre-service preparation scale with missing values, the descriptive statistics were analyzed for each question after missing values were removed, with the sample size, mean, standard deviation, skewness, and kurtosis reported (Table 2). Of the 10 questions on the pre-service preparation scale, Q1 regarding "Knowledge and understanding of the subject taught" has the highest mean value, which indicates to some extent that teachers have good subject knowledge and understanding before entering the profession. Q6 regarding "Interdisciplinary skill teaching" (e.g., STEAM, critical thinking, problem-solving, etc.) has the smallest mean value, explaining that the PSTP for teaching capabilities for interdisciplinary skills needs to be improved. It could also be observed that the mean values of Q1 to Q4 are significantly larger than those of Q5 to Q10. There may be differences in the latent variables measured in the two parts of the questions that require further analysis.


**Table 2.** Descriptive Statistical Analysis of the Pre-Service Teacher Preparation Scale.

#### *3.2. Exploratory Factor Analysis of Pre-service Teacher Preparation*

It may be biased that a subjective choice is made to split a full scale into subscales given the large number of components included in the PSTP scale. Consequently, we used the exploratory factor analysis approach to analyze the PSTP scale. Through the principal component analysis (PCA), we could know that the overall KMO (Kaiser–Meyer–Olkin) value of the scale was equal to 0.924, which was greater than 0.7. The *p*-value of Bartlett's sphericity test was less than 0.05, indicating that the information overlapping between questions was high and suitable for the factor analysis. Table 3 reports the eigenvalues and variance contribution rate of the factor analysis in which the eigenvalues of common factor 1 and common factor 2 are greater than 1, and the eigenvalues of the remaining common factors are less than 1. Meanwhile, the variance contribution rates of common factor 1 and common factor 2 are 65.015% and 12.901%, respectively, indicating that two common factors are extracted to replace 77.916% of the information of the original scale. By analyzing the eigenvalues and the variance contribution rates, it could be confirmed that the scale was suitable for extracting two common factors with the pre-service preparation scale split into two subscales.


**Table 3.** Eigenvalues and Variance Contribution Rates of the Pre-service Teacher Preparation Scale.

The factor load array was rotated using the Kaiser standardized maximum variance method to further determine the measured question items for each subscale. The rotated factor load array is shown in Table 4. For the rotated factor load array, the main focus was on the loads of each item by the factors and the larger loads could be grouped under the common factor. By analyzing the factor load of each question, it could be found that loads of Q1 to Q4 were large by common factor 1 and the factor loads of all questions were greater than 0.6. According to the connotation covered by the common factors, we could name common factor 1 as content knowledge. Loads of Q5 to Q10 were large by common factor 2 and the factor loads of all questions were greater than 0.6. We could name common factor 2 as pedagogical content knowledge according to Table 1.


**Table 4.** Rotated Factor Load Array for Pre-service Teacher Preparation Scale.

The Cronbach's alpha for the content knowledge subscale was 0.943 by further calculating the scale reliability. The Cronbach's alpha for the pedagogical content knowledge scale was 0.921. The reliability of the two subscales was good.

Q4 0.852 0.384 Q5 0.437 0.747 Q6 0.219 0.808 Q7 0.201 0.753 Q8 0.463 0.717 Q9 0.331 0.831 Q10 0.281 0.844

#### *3.3. Descriptive Statistical Analysis of Content Knowledge and Pedagogical Content Knowledge*

Table 5 reports the sample size (N), mean (Mean), standard deviation (St. Dev), minimum (min), maximum (max), skewness, and kurtosis for both content knowledge and pedagogical content knowledge. For the overall sample, the sample size was 501 teachers. The mean of content knowledge in the two PSTP subscales was 14.886, which was higher

than the median, while that of pedagogical content knowledge was 17.501, which was less than the median. It indicates that teachers were better prepared for knowledge in the PSTP self-assessment, while their preparation for teaching ability was less adequate than content knowledge. Likewise, the standard deviation of pedagogical content knowledge was large, reflecting the high discrete of the pedagogical content knowledge of the sample teachers.


**Table 5.** Descriptive Statistical Analysis of Teacher Behavior and School Support.

The results of the descriptive statistics were further analyzed in three subjects: Chinese, mathematics, and chemistry. In the sample of Chinese teachers, the valid sample size for the pre-service preparation scale was 178, and the Chinese teachers' content knowledge and pedagogical content knowledge were below the overall level. In the sample of mathematics teachers, with a valid sample size of 199 for the pre-service preparation scale, both teacher content knowledge and pedagogical content knowledge in mathematics were above average. In the sample of chemistry teachers, with a valid sample size of 124 for the pre-service preparation scale, the chemistry teachers' content knowledge and pedagogical content knowledge were lower than the overall level but higher than those of the Chinese teachers.

By subject, the results of the descriptive analysis revealed differences in teacher behaviors. If not by subject, the overall impact of various teacher behaviors on student performance may be estimated with biased results.

## *3.4. Effects of Pre-Service Teacher Preparation on Student Performance by Subject*

After controlling for individual students' characteristics, school fixed effect, and the teachers' characteristic variables (including gender, academic qualification, whether they graduated from normal universities, whether they were holding officially approved positions/bianzhi, years of teaching, and professional ranks), this study estimated the PSTP effects on student performance using multilayer linear regression, as shown in Table 6.

Regarding Chinese, column (1) shows the PSTP effects on student performance, and content knowledge has a significantly negative effect on student performance (*p* < 0.05). Each standard deviation increase in the content knowledge of Chinese teachers is associated with a significant decrease of 0.043 standard deviations in student performance. Pedagogical content knowledge has a significantly positive effect on student performance (*p* < 0.05). For every standard deviation increase in Chinese teachers' pedagogical content knowledge, the student performance is significantly enhanced by 0.053 standard deviations. In the case of Chinese, teachers' content knowledge related to subject knowledge, pedagogy, and teaching methods learned in their pre-graduation specialized programs do not contribute to student performance, even in reverse. In contrast, teachers' pre-graduation pedagogical content knowledge concerning interdisciplinary skills teaching, applied information technology teaching, and student development and assessment significantly contribute to student performance. Considering that 91.3% of the sample Chinese teachers graduated

from normal schools, normal universities should focus more on Chinese-related prospective teachers' pedagogical content knowledge, such as interdisciplinary skill teaching in training.


**Table 6.** Effects of Teacher Behavior on Student Performance.

Note: (1) standard deviations are in parentheses; (2) \*, \*\*, and \*\*\* represent 10%, 5%, and 1% in significance, respectively; (3) teachers' characteristic variables include gender, academic qualification, whether they graduated from normal universities, whether they were holding officially approved positions/bianzhi, years of teaching, and professional ranks; and (4) liberal arts and sciences indicate the discipline chosen by students, and the graduation year is 2016–2019.

With respect to mathematics, column (2) shows the PSTP effects on student performance and both content knowledge and pedagogical content knowledge have a negative but not significant effect on performance. Thus, PSTP in mathematics does not directly significantly influence students' maths performance.

For chemistry, column (3) embodies the PSTP effects on student performance, and content knowledge has a positive but insignificant effect. Pedagogical content knowledge has a significantly negative effect on student performance (*p* < 0.1). For every standard deviation increase in the teachers' pedagogical content knowledge, the student performance in chemistry will decrease by 0.05 standard deviations. In the case of chemistry, teachers' pre-graduation pedagogical content knowledge concerning interdisciplinary skills teaching, applied information technology teaching, and student development and assessment have a significant inverse effect on student performance.
