Students’ and teachers’ critical thinking in science education: are they related to each other and with physics achievement?

ABSTRACT

Background

Although there is an increasing recognition of the importance of critical thinking in science education across the world, the relevant research literature is extremely thin. There is an urgent need to generate more empirical evidence.

Purpose

We examined two essential issues concerning critical thinking (the relationship between students’ and teachers’ critical thinking as well as the relationship between critical thinking and science achievement) in the context of physics education.

Sample

We utilized data from the 2017 (Chinese) Program for Regional Assessment of Basic Education Quality with 46,820 students under 547 teachers from 128 schools.

Design and methods

We developed multiple membership multilevel models to account for students from different schools and under multiple teachers (in each school).

Results

After control over student, teacher, and school characteristics, although Chinese eighth graders demonstrated statistically significant critical thinking, there was not any statistically significant relationship between students’ and teachers’ critical thinking. With the same control, there was a statistically significant relationship between students’ critical thinking and students’ physics achievement among Chinese eighth graders. Meanwhile, there was not any statistically significant relationship between teachers’ critical thinking and students’ physics achievement (over and above the relationship between students’ critical thinking and students’ physics achievement).

Conclusion

We suggest that critical thinking may need to become an essential and explicit outcome of science education to help establish a relationship between students’ and teachers’ critical thinking and promote the correlation between critical thinking of students and teachers and students’ science achievement.

KEYWORDS:

• Students’ critical thinking
• teachers’ critical thinking
• physics achievement
• multiple membership multilevel model

Acknowledgments

We acknowledge the all-round support from the Regional Assessment of Education Quality (RAEQ) program of the Collaborative Innovation Center of Assessment toward Basic Education Quality at Beijing Normal University, especially for the specialized statistical support from the team led by Prof. Jian Liu and Prof. Hongyun Liu.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. The literature meanwhile contains further limited evidence on correlations of critical thinking with other aspects than academic achievement. This is particularly true concerning self and metacognition measures. The reason could be that these measures are often considered a part of critical thinking (dispositions). For example, metacognition is considered as critical awareness and reflection of a person’s own thinking process (e.g. Hanley 1995). Self-confidence is another example (e.g., Kavenuke, Kinyota, and Kayombo 2020). When researchers do connect critical thinking with self-confidence, there is some general belief that they are correlated (see West, 2018) as well as some empirical evidence of both positive and negative correlations among nursing students (e.g., Brown and Chronister 2009; Hoffman and Elwin 2004). In addition, there is some empirical evidence of critical thinking in a positive correlation with self-esteem and a negative correlation with state anxiety among nursing students (e.g., Pilevarzadeh et al. 2014; Suliman and Halabi 2007). Most empirical evidence is limited to the field of nursing largely because of their popular clinical judgment model (Tanner, 2006) in which critical thinking, clinical judgment, decision making, and problem solving are all interchangeable.

2. School contextual characteristics were not controlled because of the use of this special sample in PRABEQ. Schools all come from the same city with similar contextual characteristics. From the perspective of school context, these schools differ mainly in terms of per student expenditure that we controlled in data analysis.

3. MMMM is estimated by means of the Markov chain Monte Carlo (MCMC) method. We used a burn-in of 1000 iterations followed by a monitoring chain of 90,000 iterations to ensure that the algorithm both converges to a set of equilibrium posterior distributions and fully explores these distributions (see Leckie and Owen 2013). With these specifications, our effective sample size (ESS) was larger than 1000 for all parameters, resulting in rather reliable MMMM estimates.

4. Before ending the description of our research method, we discuss the issue of missing values. Missing data were 0.06% on age and 8.26% on SES at the student level. At the teacher level, because each variable was a weighted average of teachers who taught one student, there were no missing values. Missing data were 9.73% on expenditure per student and 0.66% on principal instructional leadership at the school level. MLwiN is able to handle missing data via multiple imputation at any level (Rasbash et al. 2005). As a result, all available data were utilized in the present analysis. Overall, giving the slight rates of missing data (at student and school levels), the capability of MLwiN to keep missing data, and the huge size of our sample, we would not expect missing data to create any serious impact on our MMMM analyses.

5. To reduce the size of Table 3, an intermediate model between Models B1 and B2 is not presented in the table. This intermediate model had the same structure as Model B2 expect that we allowed the relationship between students’ critical thinking and students’ physics achievement to vary across teachers and schools. Variances were trivial, and the absolute difference in DIC (about 235) between this intermediate model and Model B2 was in favor of the latter. As a result, we chose Model B2 for the table.

6. Again, to reduce the size of Table 3, an intermediate model between Models B2 and B3 is not presented in the table. This intermediate model had the same structure as Model B3 expect that we allowed the relationship between teachers’ critical thinking and students’ physics achievement to vary across schools. Variance was trivial, and the absolute difference in DIC between this intermediate model and Model B3 was about 3 (i.e. the two models fit the data equally well). As a result, we chose the simpler Model B3 for the table.

7. As mentioned earlier, the present analysis utilized the special sample in PRAEBQ (the general sample does not have teacher data). We did not find any statistically significant results except teaching experience at the teacher level. We could then ignore the teacher level and apply the same (student and school) model to data from the general sample to see if our main findings still hold. We did not pursue because of the results from the present analysis. To ignore the teacher level may not be appropriate due to three reasons. First, there was statistically significant variance in physics achievement at the teacher level. In multilevel modeling, a level is sometimes ignored but only when there is not any statistically significant variance at that level. Second, the teacher level was necessary to achieve a better model-data-fit result. This point was illustrated in our discussion on Models B3 and B4. Finally, the function of the teacher level to adjust over teacher effects has been theoretically established (see Bedasse 2017), indicating that teacher data need to be used when available in any model that examines students’ schooling outcomes. Overall, in the context of the present analysis, the appropriateness of a simplified two-level model with students nested within schools can be questionable. In other words, such a model would be subjected to considerable errors in model specification.

8. In our (final) MMMM, apart from the examination on the relationship between critical thinking and science achievement, we were also in a position to examine school effects on science achievement. As a common practice of multilevel modeling, we would start with the null model (Model B1) on variance distribution across students, teachers, and schools. We found statistically significant variance among schools, which was expected given that there are ranks of schools with very different student intakes in China. But we did not find any statistically significant school effects (see Appendix E). Thus, there was variance in science achievement among schools, but our school-level variables were not able to account for this variance. Such a pattern can be reasonable in China given the strong societal pressure on schools to produce competitive graduates (e.g. for higher education). School principals work hard in response, but the existence of ranks makes it possible that the same hardworking principals work in schools of different ranks. As a result, there is little correlation between school-level variables and science achievement. We offer this explanation as a hypothesis that unfortunately could not be tested with our data in the present analysis.