Bilişsel tanı modellerinde üst düzey düşünme becerilerinin ölçülmesinde ikili ve çoklu q matris yapılandırmasının karşılaştırılması
Küçük Resim Yok
Tarih
2022
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Ege Üniversitesi
Erişim Hakkı
info:eu-repo/semantics/openAccess
Özet
Bu tezde, "Uluslararası Geniş Ölçekli Sınavlarda Türkiye'nin Matematik Başarısını Arttırabilmek İçin Bir Model Önerisi: Bilişsel Tanıya Dayalı İzleme Modelinin Etkililiği." 115K531 numaralı TÜBİTAK projesi kapsamında elde edilen son test verileri kullanılmıştır. Araştırma için seçkisiz olarak seçilmiş 3705 öğrenciden oluşan örneklem kullanılmıştır. Kullanılan veri seti 115K531 numaralı TÜBİTAK projesi kapsamında ve bu tezde kullanılan 30 adet matematik sorusuna cevap veren öğrencileri içermektedir. Araştırma kapsamında üst düzey düşünme becerilerine yönelik sınıflandırmalar DINA ve pG-DINA modeller ile yapılmış ve Q matrisin hangi tür yapılandırma ile üst düzey düşünme becerilerini ölçmeye daha uygun olduğu saptanmaya çalışılmıştır. Çoklu Q matris düzenlemesi düzeyler 0-1-2-3 olmak üzere dört düzeyli olarak oluşturulmuştur. Araştırma kapsamında DINA ve pG-DINA modellerin becerileri ne düzeyde uyumlu sınıflayabildiğine yönelik bulgular, bütün becerilerin yaklaşık %80 oranında uyumlu sınıflandığını işaret etmektedir. Modellerin gerçek olasılıklar ve sonsal olasılıklar noktasında kendi içlerindeki uyumu incelendiğinde pG-DINA model DINA modeldeki bazı uyumsuzlukları göstermemektedir. Göreli uyum indeksleri daha küçük değerler alarak pG-DINA modelin daha uyumlu ve kullanışlı olduğuna işaret etmektedir. DINA ve pG-DINA modellerin parametreleri karşılaştırıldığında pG-DINA modelin parametre ortalama değerlerinin DINA modele göre düşük olduğu ve birçok parametre değerinin modeller arasında farklılık gösterdiği bulunmuştur. Klasik test teorisi bağlamında toplam puanlar ve z puanları incelendiğin hem DINA model hem de pG-DINA model örtük sınıfları özellik sayı ve düzeyi artışında tutarlı bir yükseliş göstermektedir. Bütün bu bilgiler ışığında pG-DINA modelin daha uyumlu olması ve daha detaylı bir tasnifleme ve tanımlama imkânı tanıması modeli özellikle pratikteki problemleri çözüldükten sonra kullanılabilir kılmaktadır
INTRODUCTION In literature, there are many definitions for higher-order thinking skills and the debates are ongoing. Still, there are some common points that define higher-order thinking skills like changing and transforming the existing knowledge, making original contributions, planning, and designing systems for various researchers. In the measurement of higher-order thinking skills, CDM classifications may be very helpful and inspiring. In this thesis, a dichotomous Q matrix and a polytomous Q matrix were modeled and the participants were classified with the appropriate models, DINA and pG-DINA. In this thesis, the main aim is to decide which CDM model can classify the participants in a more useful way and a better fit. For this study, the post-test data of the TUBITAK project "A Recommended Model to Increase Success Level of Turkey in Mathematics in International Wide-Scale Exams. Effectiveness of Cognitive Diagnosis Based Tracking Model" which numbered 115K531 was used. METHODOLOGY The research sample consists of 3705 students selected by random sampling method. The data set consists of the answers of the participants to a 30-question long mathematics test. DINA and pG-DINA models classified participants in three defined higher-order thinking skills in Q matrices. Polytomous Q matrix consists of four levels of each skill which are 0-1-2-3. The fit percentage CDM classifications for measured skills, the fit between posterior probabilities and true probabilities, relative fit indices, guess and slip parameters of the models and mean correct answers with z points of each latent class were analyzed to assess DINA and pG-DINA model in terms of usefulness. FINDINGS It has been observed that the fit of classification for each skill is approximately around %80 between DINA and pG-DINA models. True probabilities and posterior probabilities for each latent class are in a fit except for DINA class "000" and "001". DINA class tended to classify more for "001" than true probabilities and "000" less than true probabilities. Relative fit indices show consistently to each other that the pG-DINA model is a more useful model than the DINA model for measuring higher-order thinking skills. Guess and slip parameters between models are significantly different for most of the items according to z analysis made with z values of 1,64; 2,33 and 3,61 that shows significance levels of .05; .01 and .001. Mean of both parameters are lower in the pG-DINA model. In the perspective of classical test theory, total correct response means, and z value means are consistently rises as the number and levels of attributes increases. There are some uncertainties about some of the pG-DINA class hierarchical order, for those examples, it is quite hard to make a solid interpretation of classical test theory points. CONCLUSIONS In conclusion, for the DINA and the pG-DINA models, the probabilities of classifying attribute existence and attribute absence are similar. The DINA model showed some misfit to classify participants who has only the attribute "analytical reasoning" aka. latent class "001" and participants with no attribute aka. latent class "000" while the pG-DINA model showed not such a big misfit between true probabilities and posterior probabilities. Relative fit indices clearly and consistently point out that the pG-DINA is a more useful model to classify higher-order thinking skills. The parameters were significantly different for most of the items unexpectedly and pG-DINA models had lower values of guess and slip parameters. For both models, the latent classes were consistently have rising correct response and z score means as the number and level of attributes increases. To sum it all, the pG-DINA model showed more fit in analyzes and has already a potential of more informative classifications. It can be concluded that pG-DINA is a better model to use classification participants for their higher-order thinking skills.
INTRODUCTION In literature, there are many definitions for higher-order thinking skills and the debates are ongoing. Still, there are some common points that define higher-order thinking skills like changing and transforming the existing knowledge, making original contributions, planning, and designing systems for various researchers. In the measurement of higher-order thinking skills, CDM classifications may be very helpful and inspiring. In this thesis, a dichotomous Q matrix and a polytomous Q matrix were modeled and the participants were classified with the appropriate models, DINA and pG-DINA. In this thesis, the main aim is to decide which CDM model can classify the participants in a more useful way and a better fit. For this study, the post-test data of the TUBITAK project "A Recommended Model to Increase Success Level of Turkey in Mathematics in International Wide-Scale Exams. Effectiveness of Cognitive Diagnosis Based Tracking Model" which numbered 115K531 was used. METHODOLOGY The research sample consists of 3705 students selected by random sampling method. The data set consists of the answers of the participants to a 30-question long mathematics test. DINA and pG-DINA models classified participants in three defined higher-order thinking skills in Q matrices. Polytomous Q matrix consists of four levels of each skill which are 0-1-2-3. The fit percentage CDM classifications for measured skills, the fit between posterior probabilities and true probabilities, relative fit indices, guess and slip parameters of the models and mean correct answers with z points of each latent class were analyzed to assess DINA and pG-DINA model in terms of usefulness. FINDINGS It has been observed that the fit of classification for each skill is approximately around %80 between DINA and pG-DINA models. True probabilities and posterior probabilities for each latent class are in a fit except for DINA class "000" and "001". DINA class tended to classify more for "001" than true probabilities and "000" less than true probabilities. Relative fit indices show consistently to each other that the pG-DINA model is a more useful model than the DINA model for measuring higher-order thinking skills. Guess and slip parameters between models are significantly different for most of the items according to z analysis made with z values of 1,64; 2,33 and 3,61 that shows significance levels of .05; .01 and .001. Mean of both parameters are lower in the pG-DINA model. In the perspective of classical test theory, total correct response means, and z value means are consistently rises as the number and levels of attributes increases. There are some uncertainties about some of the pG-DINA class hierarchical order, for those examples, it is quite hard to make a solid interpretation of classical test theory points. CONCLUSIONS In conclusion, for the DINA and the pG-DINA models, the probabilities of classifying attribute existence and attribute absence are similar. The DINA model showed some misfit to classify participants who has only the attribute "analytical reasoning" aka. latent class "001" and participants with no attribute aka. latent class "000" while the pG-DINA model showed not such a big misfit between true probabilities and posterior probabilities. Relative fit indices clearly and consistently point out that the pG-DINA is a more useful model to classify higher-order thinking skills. The parameters were significantly different for most of the items unexpectedly and pG-DINA models had lower values of guess and slip parameters. For both models, the latent classes were consistently have rising correct response and z score means as the number and level of attributes increases. To sum it all, the pG-DINA model showed more fit in analyzes and has already a potential of more informative classifications. It can be concluded that pG-DINA is a better model to use classification participants for their higher-order thinking skills.
Açıklama
Anahtar Kelimeler
Eğitim ve Öğretim, Education and Training, Psikoloji, Psychology, İstatistik, Statistics, Bilişsel tanı modelleri, Cognitive diagnostic models, Ölçme-değerlendirme, Measurement and evaluation