Teaching strategy specialization and student achievement

ABSTRACT

This paper aims to provide evidence on whether the specialization of schools in certain teaching strategies contributes to promoting student skills. Specifically, we will focus on comparing those that make intensive use of innovative practices with those specialized in the use of traditional methodologies. By employing propensity score matching (PSM) to reduce potential bias related to the different characteristics of schools, we provide robust evidence that specialization in the use of innovative teaching practices does not lead to better academic performance and may even be harmful to some competencies.

KEYWORDS:

• Teaching strategies
• educational evaluation
• student achievement
• propensity score matching
• education policy

Acknowledgements

The authors would like to express their gratitude to Carmen Tovar and all the personnel working at the Spanish National Institute for Education Evaluation for their technical support.

Disclosure statement

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

Notes

1 We benefit from the fact that this is the last wave conducted for this survey in Spain, using very specific information not available in any other large-scale assessment with these characteristics in Spain.

2 See Cordero, Cristóbal, and Santín (2018) for comprehensive review of empirical studies applying quasi-experimental techniques to data from these three databases.

3 For a detailed description of this approach, see Wu (2005).

4 The use of a single plausible value does not make a noticeable difference within large enough samples as stated in several technical reports on international databases (for example, OECD 2009, 44).

5 The resulting values are standardized so that both indices are ranged in the same scale.

6 This is a logical assumption when working at the secondary education level, where, unlike primary education, students have different teachers for each subject.

7 We established a percentage of 25% rather than a lower proportion (20, 10 or 5%) because we need to have enough schools representing each teaching style to be able to later implement propensity score matching.

8 Although this is the general criterion used to classify schools into two groups, in the results section we present an alternative classification as a robustness check. Specifically, we compare the top quartile of the two distributions of results with the two bottom quartiles (bottom 50%) to ensure enough difference in the treatment between the two groups.

9 This indicator is derived from responses to a question rated on a seven-level scale: (1) less than 5 years; (2) between 5 and 9; (3) between 10 and 14; (4) between 15 and 19; (5) between 20 and 24; (6) between 25 and 29 and (7) 30 or more.

10 These two variables are defined considering teacher responses to two questions rated on a four-point Likert-type frequency scale.

11 This indicator was constructed following an analogous procedure to the method used by OECD experts to calculate the PISA socioeconomic index (ESCS), that is, using a factorial analysis to synthesize the information on variables related to parents’ educational level, parent’s occupational level and cultural possessions (mainly represented by the number of books at home). These three factors are widely recognized in the literature as the ones that best represent family socioeconomic status (Yang and Gustafsson 2004).

12 This approach avoids biases arising from the fact that the values of school-level aggregate variables are correlated for students belonging to the same school (Hox 2010).

13 It is worth mentioning that some authors like Altonji, Elder, and Taber (2008) or Oster (2019) proposed alternative methods to assess potential bias from unobservable variables. However, those methods are not suitable in our framework, as they are applicable to linear models.

14 See Guo and Fraser (2010) for details.

15 Kernel methods can be defined as non-parametric matching estimators that compare the result of each treated unit with a weighted mean of the results of all the comparison group units using the highest weightings for units whose propensity score is closer to the comparison unit value.

16 In the estimated regressions we include all the covariates defined in the previous section as explanatory variables. However, to save space, we only show the estimated parameters for the student-level variables and the two school-level dummy variables representing teaching specialization in a certain teaching style (classical or innovative).

17 Estimates are bootstrapped by cluster (schools) using 50 replications to calculate approximate standard errors.

18 These estimates have been made including all the control variables and the clustered structure of students, but we only show the value of the estimated parameter of interest to save space.

Additional information

Funding

The authors are supported by Junta de Extremadura and ‘ERDF A way of making Europe’, through the Project GR21049.