The Effect of Different Critical Thinking Teaching Approaches on Critical Thinking Skills: A Meta-Analysis Study

Yayın Tarihi: 25.09.2019 Abstract Critical thinking (CT) which is self-regulatory thinking process includes some skills like interpretation, analysis, evaluation, inference and evidence questioning. There are skill-based, content-based and mix CT teaching approaches in literature. It is important to find out how effective these approaches are on CT skills rather than whether they are effective on CT. So this study aimed to determine to what extent the content-based and skill-based teaching of CT is effective in improving the CT skills. 21 results of 17 research studies were included in the meta-analysis. The content-based teaching of CT is strongly effective in improving students’ CT skills. This effect level does not differ significantly by whether the sample group is gifted or normal students and by educational level of sample groups. Also, the skill-based teaching of CT is strongly effective in improving students’ CT skills. Besides, the effect size of the studies which applied the skill-based teaching of CT is higher than the effect size of the studies which applied the content-based teaching of CT. However, there is no significant difference between these two approaches in terms of improving CT skills. So skill based and content based critical thinking teaching approaches can be used for all students from different education levels to improve


Introduction
Critical thinking (CT), which dates back to Socrates (Vandenberg, 2009), is still a subject of considerable interest to cognitive theorists and educators today (Akınoğlu, 2001;Ay and Akgöl, 2008). CT is defined as a conscious and self-regulatory thinking process that includes skills such as interpretation, analysis, evaluation, inference and evidence questioning (Facione, 1990). According to Watson-Glaser (2010), the CT process involves the ability to recognize the existence of a problem, to question the source of information and to question the accuracy of the evidence, and to evaluate different data and evidence. In other words, the source of the CT process can be regarded as achieving results with observations and knowledge that have been questioned and tested for accuracy (Paul, 1992). Therefore, with the CT, individual acquires the skill of thinking properly and having the right information about their environment (Schafersman, 1991). Because CT skill is a reflective and logical thinking process (Ennis, 1985) in which the individual seeks solid evidence when deciding how to act or what to believe (Schafersman, 1991;Meltzoff and Cooper, 2018). As a result of this act of thinking, one also takes control of the structures in their own thinking process and develops them according to the intellectual standards they set themselves (Scriven and Paul, 2004;Paul and Elder, 2019).
Some different approaches have been adopted by different researchers about the teaching of CT (Beyer, 1987;Resnick, 1987;Lipman, 1988;Kenyon and Guillaume, 2014). Prior to previous studies in literature, it seems that three basic approaches are widely used in the teaching of CT skills. These approaches are content-based teaching of CT, skill-based teaching of CT, and mixed teaching of CT.
According to the first of these approaches, CT skills are taught by integrating them into the content of a course. According to McPeck (2016), it is meaningless to teach CT skills independently of a context as the individual thinks on content or subject when using thinking skills. There is no set of CT skills that can fit all contents or topics. Therefore, the skills in question need to be integrated into educational programs and taught in all courses (Resnick, 1987). In this way, students will be able to use their skills in real life and in different courses.
The skill-based approach in the teaching of CT argues that CT skills should be taught as a different course independent of the content of the courses in the curriculum (Sternberg, 1985). According to Ennis (1991) who identified the 12 dimensions of CT, all of these dimensions can be taught, and can be acquired by students. Therefore, there is no problem in the teaching of these skills in a separate course on a skill basis (Ennis, 1991). When CT skills are taught on the content-based basis within a course, the focus is on the course content and CT skills are ignored (Lipman, 1988). Hence, the skill-based teaching approach, which allows to avoid focusing on the subjects in the course content and repeating these subjects all the time, is more effective in bringing CT skills and using these skills in other courses (Ennis, 1991;Beyer, 1991). In the mixed teaching approach, the teaching of CT starts on the skill basis, and then, continues on the content basis (Perkins and Salomon, 1989). Therefore, advocates of each approach mention different advantages or disadvantages of CT. With this information at hand, one cannot be sure which approach creates more effective results in the teaching of CT.
As for the studies investigating the effect of the content-based and skills-based teaching approaches on CT skills in Turkey, the question which approach is more effective in promoting the CT skills cannot be answered because teaching approaches of CT used in all of these studies conducted independently by different researchers and at different times significantly enhanced the CT skill. However, it is difficult to identify to what extent the approach is effective in improving CT skills. It is more important to find out how effective this approach is on CT skills rather than whether it is effective on CT. Meta-analysis studies that combine the results of similar studies conducted on a certain subject by different people are of importance (McMillan and Schumacher, 2001) as these studies allow a consistent interpretation of the information accumulated in a specific area (Akgöz, Ercan and Kan, 2004). The meta-analysis method aims to bring together the results of the studies in the literature in a consistent and coherent manner with the statistical methods (Cohen, 1988;Chambers, 2004), to discuss the results of these quantitative studies in a holistic manner (Creswell, 2014) and to achieve more extensive and generalizable results, gaining an upper point of view (Erkuş, 2009).
Although there are several studies on CT in the literature in Turkey, a meta-analysis study that is able to answer this question is yet to be carried out. In the international literature, however, there are studies investigating the experimental studies which have been conducted on CT skills in a systematic way (Abrami et al., 2008;Behar-Horenstein and Niu, 2011;Abrami et al., 2015). In this context, the question "How effective are the content-based and skill-based teaching approaches of CT in improving CT skills?" presents the questions of this research to fill this gap. Thus, a broader perspective will be taken to see experimental studies in Turkey using these teaching approaches, and a general consideration will be achieved on which of the approaches are more effective in Turkey. It is anticipated that such consideration will guide teachers, academics and other researchers in their own work.
To this end, answers to the following questions were sought for: 1. At what level does the content-based teaching approach of CT affect CT skills? a. Does this level of effect vary by different variables (whether the sample is normal or gifted students and sample's educational level)?
2. At what level does the skill-based teaching approach of CT affect CT skills?
3. Is there any difference between the effect levels of the content-based and skill-based teaching approaches of CT on CT skills?

Research Model
Meta-analysis method was used in this study which aimed to determine the effects of contentbased and skill-based teaching approaches of CT on CT skills. Meta-analysis is a method that helps gather, combine findings from similar studies conducted on a given subject at different places and times and calculate a shared effect size through statistical methods (Cohen, Manion & Morrison, 2007). Thus, a common conclusion can be drawn about these studies, and it becomes possible to make a general interpretation (Hedges & Olkin, 1985). The steps followed in this study can be listed as follows: 1. Identify the problem 2. Review the literature for the collection of studies 3. Coding of the studies 4. Data analysis and interpretation The problem of this study is to determine the effect size of the content-based and skillbased approaches used in the teaching of CT on CT skills.

Collection of Studies
Inclusion criteria were determined to decide whether to include studies collected in the literature review in the first place. Then, the suitability of the studies to the inclusion criteria was examined by two different people, and the studies that met these criteria were included in the analysis.
The criteria used for the inclusion of the studies that were found in the literature review can be listed as follows: 1. Studies conducted between 2005 and 2019 in Turkey, 2. Studies written in Turkish or English, 3. Studies that are postgraduate theses or papers published in peer-reviewed scientific journals, 4. Studies conducted with experimental method, 5. Studies that implemented the content-based or skill-based approaches of CT in the experimental group and traditional teaching approach in the control group. 6. Studies that used instruments measuring the CT performance. In other words, studies using data collection instruments developed to determine the CT disposition were not included in the study. 7. Studies that clearly specify the statistical data required to calculate the effect size.
The studies included in the research in consideration of the abovementioned criteria are the postgraduate theses and papers conducted on the teaching of CT in Turkey between 2006 and 2018. According to Wells and Littel (2009), one of the criticisms against meta-analysis studies is about the quality of primary studies included in the analysis. Therefore, only the postgraduate theses and the papers published in peer-reviewed scientific journals were included in this study and the papers on the teaching of CT presented in scientific activities such as congresses or symposiums were not included to keep clear of this criticism and mitigate the quality problem.
To access the theses on the subject, a search was performed in Higher Education Council (YÖK) national thesis database with the Turkish keywords of "eleştirel düşünme öğretimi", "eleştirel düşünme", "eleştirel düşünme becerisi", "içerik temelli eleştirel düşünme öğretimi", "beceri temelli eleştirel düşünme öğretimi" and the English keywords of "CT teaching", "CT", "improving CT skills", "teaching CT" between 20.02.2019 and 01.03.2019. 103 theses were accessed in the search. The oldest thesis among them was written in 2005 and the latest in 2018. The theses were reviewed in terms of inclusion criteria. Even if the terms of skill-based or content-based teaching of CT are not directly available in the title or content of the theses, the studies that were decided to have used content-based or skill-based teaching of CT after reading the experimental procedures and the stages of the CT teaching were included in the study. It was found after reviewing the theses that 15 theses were fit for the research purpose and the inclusion criteria. Thus, 15 theses were accessed in the literature review and included in the meta-analysis. Since 2 experimental groups were formed in each three of the theses, it was possible to access the results of 18 experimental studies which were conducted with 18 experimental and 18 control groups.
Another search was made with the abovementioned keywords to access papers published on the subject on the databases of ULAKBİM, Google Scholar, EBSCOhost, Web of Science, and ERIC between 20.02.2019 and 01.13.2019, and those from Turkey were downloaded on the computer. The downloaded papers were evaluated by the inclusion criteria, and it was seen that 5 papers met the criteria. In case the postgraduate theses might have been published separately, the papers and theses were compared, and it was found that the case applied to 3 papers. Therefore, this study did not include these papers but their thesis versions. Consequently, 2 papers were found in the literature review and included in the meta-analysis. However, since one of the papers has 2 experimental groups and 1 control group, the findings of 2 experimental studies could be derived from this study, and in the end, the findings of 3 experimental studies in total were included in the study.
Then, the bibliography sections of the theses and papers were reviewed in detail to access other studies on the subject. Yet, no study meeting the inclusion criteria was observed with this method. As a result, there are 309 samples in total in the experimental groups in which the content-based teaching approach was applied, 204 samples in total in the experimental groups in which the skill-based teaching approach was applied, and 473 samples in the control groups in 15 postgraduate theses and 2 papers that meet the inclusion criteria and could be accessed.

Coding of Studies
A form was developed by the researcher to code the studies included in the meta-analysis. The coding form involves information such as name of study, its publication year, type and author(s), which teaching approach of CT it used, instrument for measuring the CT skill, field of study, its experimental and control groups; characteristics and numbers of the sample, and data required for the calculation of effect size. The content validity of the coding form was achieved through expert opinion, and small changes were made to the form as a result of the feedbacks.
The studies included in the analysis were coded with the coding form by the researcher and by another person who is doing doctorate in educational sciences to increase the reliability of the study and minimize possible errors. Using the formula (Number of Agreements/[Number of Agreements+ Number of Disagreements]) proposed by Miles and Huberman (1994), the coefficient of concordance among the codes were calculated and found 0.98. According to Miles & Huberman (1994), the coefficient of concordance above 0.70 indicates a reliable study. So it can be concluded there is a high concordance between the two coders. The characteristics of the studies included in the meta-analysis can be seen in Table 1. According to Table 1, most of the studies included in the meta-analysis (n:9) are doctoral theses. The doctoral theses are followed by postgraduate theses (n:6) and doctoral theses (n:2), respectively. In addition, the studies were most conducted in 2014 (n:3), 2011 (n:3), 2013 (n:2) and 2009 (n:2) respectively.

Data Analysis and Interpretation
In this study aiming to determine at what level the content-based and skill-based teaching approaches of CT affect CT skills, these teaching approaches of CT were set as independent variables while CT skills was decided to be the dependent variable.
The data obtained on the studies with the coding form were transferred to the CMA software. The arithmetic mean, standard deviation and sample number of 21 experimental studies were utilized for meta-analysis. The coding form used for transferring the data to the software prevented incorrect data entries.
In this study, the p value was checked to decide whether there was heterogeneity in the first place, then the Q value was checked according to the value in the X 2 table, and finally the I 2 value was checked.
Hedge's g coefficient was utilized in the effect size calculations, and the confidence level was determined to be 95% in all calculations in the study. The classification introduced by Cohen, Manion & Morrison (2007) was used for the interpretation of the effect size. The classification of effect sizes is as follows: The criterion values shown in Table 2 were used to interpret the effect size achieved in the study. The funnel plot was used to determine whether there was publication bias. Also, Rosenthal's Fail-Safe N test was used to support the findings of funnel plot.

Checking Studies Which Applied Content-Based Teaching of CT for Publication Bias
It was tested before calculating the effect sizes whether there was publication bias. The results of funnel plot which allowed us to comment on the presence/absence of publication bias are as follows:

Figure 1. Funnel plot of effect sizes
To be able to conclude that there is no publication bias, effect sizes of studies need to be distributed symmetrically around the general effect size in the funnel plot (Borenstein et al., 2009). Hence, according to the funnel plot in Figure 1, the effect sizes of 16 studies included in the research are symmetrically distributed around the general effect size except one study. Therefore, it is possible to conclude that there is no publication bias. In addition, Rosenthal's Fail-Safe N test was performed for reinforcing the finding achieved in the funnel plot. The findings are presented below:  Table 3 indicates that the findings achieved in Rosenthal's Fail-Safe N test coincide with the funnel plot findings. It was concluded in Rosenthal's Fail-Safe N test that 657 studies with an effect size of zero need to be included further in the analysis for the meta-analysis results achieved in this study to lose their significance. It can be inferred from this number which is much higher than the number of reviewed studies that there is no publication bias (Rosenthal, 1979). Besides that, Begg and Mazumdar rank correlation test is another way to test presence/absence of publication bias. As a result of Begg and Mazumdar rank correlation test, p value should be over 0.05 to say there is no publication bias (Begg & Mazumdar, 1994). So it can be said that there is no publication bias among studies that applied content-based teaching of CT (Tau b=0.27; p >0.05). Thus, the findings achieved both in the funnel plot, Rosenthal's Fail-Safe test and Begg and Mazumdar rank correlation test show that there is no publication bias in this study.

Checking Studies Which Applied Content-Based Teaching of CT for Heterogeneity
Heterogeneity test is important for a meta-analysis because the statistical model to be applied in meta-analysis is decided according to the result of this test (Huedo-Medina et al., 2006). A heterogeneity test was accordingly performed to decide which statistical model would be used in this study. The findings of the heterogeneity test are given below:  Table 4, Q value is 28.072. This value is above the critical value of 24.996 prescribed for 15 degree of freedom and also at 95% significance level in the X 2 table. Then it is obvious that there is heterogeneity among the studies according to the Q value achieved. Nevertheless, I 2 value that is not influenced by the number of studies and can measure heterogeneity more accurately (Petticrew & Roberts, 2006) was also calculated to support the Q statistic which is likely to fall weak in identifying the heterogeneity in case of low number of studies subjected to the meta-analysis (Huedo-Medina et al., 2006). The table shows that the I 2 value is 46.5%. Cooper, Hedges and Valentine (2009) state that I 2 being 25% refers to low, being 50% to moderate, and being 75% to high heterogeneity. Hence, this value indicates that there is moderate heterogeneity among the studies. Moreover, the p value is below 0.05 (p=0.021). As a result, all values obtained (Q=28.072, p>0.05, I 2 =45.565) show that there is heterogeneity among the studies and the random effects model can be used to calculate the effect size.

Checking Studies Which Applied Skill-Based Teaching of CT for Publication Bias
The results of funnel plot which allowed us to comment on the presence/absence of publication bias are as follows:

Figure 2. Funnel plot of effect sizes
According to the funnel plot in Figure 2, effect sizes of 5 studies included in the research are symmetrically scattered around the general effect size except in two studies. It can be therefore concluded that there is no publication bias. Rosenthal's Fail-Safe N test was utilized for reinforcing the finding achieved in the funnel plot. The findings are presented below:  Table 5 shows that the findings achieved in Rosenthal's Fail-Safe N test in Table 5 coincide with the funnel plot findings. 112 studies with an effect size of zero need to be conducted further for the meta-analysis results obtained from the study on the skill-based teaching of CT to lose their significance. Also the result of Begg and Mazumdar rank correlation test supports that there is no publication bias among studies that applied skill-based teaching of CT (Tau b =0.20; p >0.05). Therefore, the findings achieved both in the funnel plot, Rosenthal's Fail-Safe test and Begg and Mazumdar rank correlation test show that there is no publication bias in this study.

Checking Studies Which Applies Skill-Based Teaching of CT for Heterogeneity
A heterogeneity test was performed first to decide which statistical model to use, and the findings are given in Table 6.

Findings Findings on the Effect Size of Content-Based Teaching of CT on CT Skills
Since heterogeneity was identified among the studies included in the meta-analysis, the effect sizes of the studies were combined using the random effects model. The general effect size obtained with the random effects model can be seen in the table below:  Table 7 shows that according to the random effects model, the general effect size of the content-based teaching of CT on CT skill is 1.111 with an error of 0.121. This is a strong effect in accordance with the classification of Cohen, Manion & Morrison (2007). Likewise, lower limit of the effect size calculated with the random effects model is 0.875, and its upper limit is 1.348 within the confidence range of 95%. The values of effect size can be assumed to be statistically significant (Z=9.198; p=0.00). In view of this finding, it can be concluded that the content-based teaching of CT strongly affects CT skills. In other words, the content-based teaching approach of CT is more effective positively and strongly on students' CT skills.
The studies were divided into two different groups to determine whether effect size differs by whether the sample group is gifted or normal students. The results of the analysis with the two groups are given in Table 8.  Table 8 shows that all effect sizes are positive, and the studies conducted on gifted students have a higher effect size (g=1.247) than the studies conducted on normal students (g=1.039). The effect sizes calculated for both study groups are strong. Furthermore, the Q value was found 0.606 in the heterogeneity test performed to determine whether the effect sizes differ by sample characteristics. This value is below the critical value of 3.841 which is prescribed for 1 degree of freedom and at significance level of 95% in the X 2 table. The achieved p value is also above 0.05 (p=0.43). It can be therefore argued that there is no significant difference between the effect sizes of the studies investigating the effect of contentbased teaching of CT on CT skills with sample groups of gifted students and normal students (Q=0.606; p=0.436). However, one can argue that the studies conducted with gifted students have a higher effect size than the studies conducted with normal students.
The studies were divided into four different groups of primary, secondary, high schools and university to determine whether the effect size differs by educational level. The results of the analysis on these four groups are shown in Table 9.  Table 9 shows that all effect sizes are positive and the studies conducted on university students have the highest effect size at 1.341. University students are followed by primary school students (g=1.172), secondary school students (g=1.151) and high school students (g= 0.704), respectively. The calculated effect size values are strong for all groups except high school students. The effect size calculated for high school students is moderate. In this case, it can be concluded that the content-based teaching of CT improves the CT skills of primary and secondary school students and university students at a strong level while improving high school students' CT skills at a moderate level. The Q value was found 4.574 in the heterogeneity test to determine whether the effect sizes differ by educational level of sample groups. This value is below the critical value of 7.185 which is prescribed for 3 degree of freedom and at significance level of 95% in the X 2 table. The achieved p value is also above 0.05 (p=0.20). It can be therefore said that the distribution is homogeneous (Q=4.574; p=0.206). In other words, educational level of sample groups is not a factor that changes the calculated effect size.

Findings on the Effect Size of Skill-Based Teaching of CT on CT Skills
Since heterogeneity was identified among the skill-based studies included in the meta-analysis, the effect sizes of the studies were combined using the random effects model. The general effect size obtained with the random effects model is given in Table 10. As seen in Table 10, the general effect size of the skill-based teaching of CT on CT skill is 1.126 with an error of 0.356 according to the random effects model. This value refers to strong effect. Likewise, lower limit of the effect size calculated with the random effects model is 0.428, and its upper limit is 1.824 within the confidence range of 95%. The values of effect size can be assumed to be statistically significant (Z=3.162; p=0.002). It can be inferred from this finding that the skill-based teaching of CT strongly affects CT skills. In other words, the skill-based teaching approach of CT is more effective strongly on students' CT skills compared to traditional teaching approaches. Table 11 presents the analysis results of the two groups formed to determine whether there is a difference between the effect sizes of the content-based and skill-based teaching approaches of CT on CT skills.  Table 11 shows that all effect sizes are positive, and the studies which used the skillbased teaching of CT have a higher effect size (g=1.126) than the studies which used the content-based teaching of CT (g=1.111). The effect sizes calculated for both groups are strong. In addition, the Q value was found 0.842 in the heterogeneity test performed to determine whether the effect sizes differ significantly. This value is below the critical value of 3.841 which is prescribed for 1 degree of freedom and at significance level of 95% in the X 2 table. The achieved p value achieved in the statistics is also above 0.05 (p=0.35). Therefore, there is no significant difference between the effect sizes of the studies investigating the effect of contentbased teaching of CT on CT skills and the studies investigating the skill-based teaching of CT on CT skills (Q= 0.842; p= 0.35). To sum up, one can conclude that students' CT skills would improve at the same rate in either content-based or skill-based teaching of CT.

Conclusion and Discussion
This study aimed to determine to what extent the content-based and skill-based teaching of CT is effective in improving the CT skill. As a result of the literature review performed to this end, 21 results of 17 research studies were included in the meta-analysis.
The first question of this study aimed to determine the effect level of the content-based teaching of CT on CT skill and whether this effect level differs by some variables. The general effect size of the content-based teaching of CT on CT skill is 1.111 with an error of 0.121 according to the random effects model. This effect value falls within the strong range. Hence, it is possible to state that the content-based teaching of CT is strongly effective in improving students' CT skills. It follows that a content-based teaching of CT can be carried out to improve students' CT skills. Aiming to re-evaluate the qualitative studies on CT skill in the literature with the meta-synthesis method, Polat (2015) concluded that CT activities integrated into in the curriculum on the content basis are highly effective in enhancing students' CT skills. Tiruneh, Verburgh, and Elen (2014) found that CT skills were significantly improved in approximately 65% of the experimental studies on the content-based teaching of CT which they reviewed. Therefore, the result that the content-based teaching of CT is strongly effective in improving the CT skills coincides with the results in the literature.
This effect level does not differ significantly by whether the sample group is gifted or normal students (Q=0.606; p= 0.436) and by educational level of sample groups (Q=4.574; p=0.206). However, although the difference is not statistically significant, one can argue that the effect size of the studies working with gifted students (g=1.247) is higher than the effect size (g=1.039) of the studies working with normal students. It is also possible to state that the studies working with university students have the highest effect size (g=1.341) among the sample groups formed by the educational level. University students are followed by primary school students (g=1.172), secondary school students (g=1.151) and high school students (g=0.704), respectively. The calculated effect size values are strong for all groups except high school students. The effect size calculated for high school students is moderate. Systematically reviewing 33 experimental studies which discussed the teaching of CT, Tiruneh, Verburgh and Elen (2014) concluded that the educational level of sample group does not significantly affect the achievement of content-based or skill-based teaching of CT in promoting the CT skills. Similarly, there are other studies showing that the educational level of sample group does not affect the success of CT teaching (Chau et al., 2001;Hitchcock, 2004). It is therefore possible to argue that the results of this study are in parallel with the results of many studies in the literature.
The second question of the study aimed to determine the effect level of skill-based teaching of CT on CT skills. According to the random effects model, the general effect size of the skill-based teaching of CT on CT skill is 1.126 with an error of 0.356. This effect value is strong. Hence, one can argue that the skill-based teaching of CT is strongly effective in improving students' CT skills. According to Bangert-Drowns and Bankert (1990), who aimed to combine and interpret the results of studies based on skill-based teaching of CT with the metaanalysis method, the skill-based teaching of CT was strongly and significantly effective in improving students' CT skills. This result achieved by Bankert-Drowns and Bankert (1990) coincides with the results obtained in this study. In parallel with the results of this research, Tiruneh, Verburgh, and Elen (2014) found that CT skills were significantly improved in approximately 80% of the experimental studies on the skill-based teaching of CT which they reviewed.
The third question of the research aimed to determine whether there is any difference between the effect levels of content-based and skill-based teaching approach of CT on CT skills? It can be argued that the effect size (g=1.126) of the studies which applied the skill-based teaching of CT is higher than the effect size (g=1.111) of the studies which applied the contentbased teaching of CT. The effect sizes calculated for both groups correspond to the strong level. In addition, it was observed that the difference was not significant in the heterogeneity test which was performed to see whether the effect sizes achieved for both approaches differ significantly (Q=0.842; p=0.35). Analyzing the experimental studies that applied CT teaching, Bankert-Drowns and Bankert (1990) concluded that there was no significant difference between the effect sizes of the studies using the content-based approach and the studies using the skillbased approach. Likewise, Behar-Horenstein and Niu (2011), who reviewed the experimental studies on CT, found that the studies which teach CT on the skill basis are more effective in improving CT skills than the studies which prefer the content-based teaching. However, they also state that this difference is not too much. Furthermore, Abrami et al. (2008) reviewed 117 studies on CT teaching with the meta-analysis method and concluded that there were no significant differences among the effect sizes of the content-based, skill-based and mixed approaches of CT even though the content-based approach had a lower effect size. In another study aiming to combine the results of 341 experimental studies that teach CT with the metaanalysis method, Abrami et al. (2015) similarly concluded that the content-based and skill-based approaches were equally effective in promoting CT skills. Examining 33 experimental studies using different CT teaching approaches, Tiruneh, Verburgh and Elen (2014) state that the studies using skills-based approach are more successful in improving CT skills than the studies using the content-based approach. However, according to Tiruneh, Verburgh and Elen (2014), this difference between the two teaching approaches is not significant. It is therefore possible to argue that the results obtained in the literature coincide with the results of this research. In addition, according to Arrington (2017), who reviewed the CT teaching in 8 public universities with methods such as interview, scale and document review, more than half of the faculty members (58.62%) working in universities think that CT should be taught on the content basis. On the other hand, the remaining group thinks that the skill-based approach is more effective in teaching CT. Therefore, it can be stated that there is not an agreement on CT teaching approaches and the number of the group that advocates the content-based and skill-based approaches is close to each other in Arrington's (2017) study. In accordance with this result, the meta-analysis through this study revealed that the two teaching approaches are not superior to each other in improving CT skills.
In short, this study found answers to the questions "To what extent is the content-based teaching approach of CT effective in improving CT skills?", "To what extent is the skill-based teaching approach of CT effective in improving CT skills?", and "Which of the teaching approaches of CT is more effective in improving CT skills?" According to the results, the content-based and skill-based approaches in CT teaching have a strong effect on the improvement of CT skills. Furthermore, there is no significant difference between these two approaches in terms of improving CT skills. It is thought that answering these questions to which more importance is ascribed fulfills the gap in the literature in Turkey and gives an idea to other researchers and practitioners. In the light of these results, the following recommendations can be provided for other researchers and practitioners:  The content-based and skill-based teaching approaches of CT can be used to promote CT skills on all education levels.
 Students can be provided with CT skills by integrating these skills into the curricula of primary, secondary and high schools on the content basis along their entire educational life.
 On university level, CT skills can be taught on the skill basis in a separate course.
 Based on the conclusion that CT skills can be strongly improved with these two approaches, in-service trainings can be given to teachers in these two approaches so that they can use them effectively in their courses.
 Limited number of experimental studies conducted with the content-based and skillbased approaches can be regarded as a limitation to this study. Hence, the meta-analysis study can be repeated and more exhaustive results can be achieved once there are more of the said studies.