Models for Square Data Stat 557 Heike Hofmann Outline • Movies • Political Views • Relationships Movie Reviews • • • Siskel/Ebert movie reviews Use expand.grid to get the data into R quickly Use a mosaicplot to plot the raw data Ebert Siskel Con Mixed Pro Con 24 8 10 Mixed 8 13 9 Pro 13 11 64 Movie Reviews • Discuss whether an assumption of marginal homogeneity holds. - draw two barcharts for a start. • get the data into a shape that you can fit the following two POLR models: count variable value 1 24 ebert con 2 8 ebert mixed 3 13 ebert pro 4 8 ebert con 5 13 ebert mixed ... library(MASS) mov.null <- polr(value~1, weight=count, data=reviews.melt) mov.main <- polr(value~1+variable, weight=count, data=reviews.melt) anova(mov.null, mov.main) • Why does this test marginal homogeneity, and what is the conclusion? Movie Reviews • • We’ve discussed symmetry and quasi-symmetry. • Using a log-linear framework fit models of symmetry and quasi-symmetry. Interpret results • If possible, compare models. Introduce a factor variable symm that makes each pair of cells (i,j) and (j,i) into one level Movie Reviews • Besides independence, there is quasi-independence (next slide) • Using a log-linear framework fit models of independence and quasi-independence. Interpret results • What is the structure between symmetry, quasisymmetry, independence and quasi-independence (if at all)? Quasi Independence Quasi-Independence Model Quasi-Independence Model Quasi-Independence Model si-Independence Model • Independence in matched pair data is usually violated because heavy diagonal usually violated (beca In matchedof pair data independence • heavily loaded) tched pair data independence usually violated (because diagonal Quasi-Independence: Quasi-independence model: fit independence for off-diago y loaded) cells on the main separately: givenmodel: off-diagonality, dodiagonal we for have independence? -independence fit independence off-diagonal cells and fit Y2 1 on the mainModel diagonal separately: log µab = λ + λY + λ Form a b + δa I (a = b) , • 1 λY a 2 λY b � �� independence log µab = λ + + + δa I (a = b) , �� with � � I is�� � �function, where an indicator main diagonal independence I is an indicator function, with with � � 1 I (a = b) = 0 � � �� � main diagonal if a = b, otherwise. Political Views • Download data polviews06.csv from our website • Discuss marginal homogeneity. • How do political views and party affiliation align? Draw a fluctuation diagram Political Views • Fit models of (quasi-)independence & (quasi) symmetry and interpret results. • Compare corresponding models. • Does number of respondents have an impact on model fit? Pre- vs Extramarital Relationships • Download data set gss-ex-pre.txt from the website • GSS data: attitude towards pre/ex-marital sexual relationships • For one of the years, fit logistic regression and check for marginal homogeneity • Does attitude change over the years?