Research on impacts of different types of special education supports is surprisingly scarce, although there are hundreds of studies on the efficacy of specific interventions implemented as a part of special education (see eg. Forness 2001, Mitchell 2014). Major challenges in research aiming to assess overall impacts of special education are at least the following: (1) special education takes different forms and types in different countries so, practices of identifying students for support vary which makes comparison of findings between countries difficult (Schwab, 2019) (2) it is difficult – for many reasons including ethical dilemmas - to find comparison groups to any students receiving special educational needs. This leaves researcher to simply to document what happened to students after receiving special education, but the essential question, what would have happened if they had not received the support, remains unanswered. The objective of this paper is to discuss possible ways out of this problem and give examples of results of different analyses utilizing the same data set. These approaches rely on longitudinal data on student support and expected outcomes as well as relevant covariates that may be related to initial decision to support the student. These analyses can be divided into two different approaches: longitudinal covariate models and matching. The first approach uses longitudinal modelling comparing outcomes for special education and other students while controlling for variables that are relevant covariates of receiving support. Two examples of this kind of analysis will be given and the first reports effects of part-time special education support on student self-concept utilizing growth curve modelling in three consecutive measurements with student between grades 5-7 (Savolainen et al. 2018). Another example with the same data is a path analysis of the effects of special education on self-concept, controlling for the effects of relevant covariates at the baseline measurement. Second approach is based on finding close enough matches for students receiving support in the data. Results of one specific type of matching using Propensity Scores is reported here. Matching requires complete data for all cases included, therefore analysis with imputed data will be used. Results from two alternative approaches for PSM with imputed data (‘within’ and ‘across’) will be reported. Results of analyses done with the different approaches aiming to answer the same research question will be compared and implications for special educational practices and research will be discussed.

Forness, S. (2001). Special education and related services: What have we learned from meta-analysis. Exceptionality, 9, 185–197. King, G., & Nielsen, R. (2019) Why Propensity scores should not be used for matching. Political analysis, 1-20, DOI: 10.1017/pan.2019.11 Mitchell, D. (2014) What really works in special end inclusive education: using evidence based teaching strategies. New York: Routledge. Morgan, P.L., Frisco, M.L., Farkas, G., Hibel, J. (2010) A Propensity score matching analysis of the effects of special education services. The Journal of Special Education, 43(4), 236-254. Savolainen, P., Timmermans, A.C., & Savolainen, H. (2018) Part-time Special Education Support Predicts the Academic Self-Concept between Grades 5 and 7. Learning and individual differences, 68, 85-95. Schwab, S. (2019). Inclusive and special education in Europe. In Sharma, U. (Ed.). Oxford Research Encyclopedia of Education. New York: Oxford University Press.