140 likes | 430 Views
Cultural Nuances, Assumptions and the Butterfly Effect: A Prelude. A Brief Introduction to Generalizability Theory for the Uninitiated. What is Generalizability Theory?.
E N D
Cultural Nuances, Assumptions and the Butterfly Effect:A Prelude A Brief Introduction to Generalizability Theory for the Uninitiated
What is Generalizability Theory? Generalizability Theory—the theory that deals with the degree to which one can say that conclusions drawn on the basis of data collected from sampling different domains can be applied with confidence to those domains.
Still Lost? What we’re talking about you best know as instrument reliability. Commonly known examples: • inter-rater reliability, • test-retest reliability, • split-halves reliability, • Kuder-Richardson reliabilities, • Hoyt reliability, and • Cronbach’s alpha
Observed scores are composed of true scores and error. The variance of the observed scores is partitioned. Then estimates are combined to produce a coefficient. Xo = Xt + Xe X=score, t=true, o=observed, e=error So2 =St2 + Se2 S2 = variance r11 = St2/So2 Test (True) Score Theory
True Score Theory and Generalizability • Generalizabilty is an extension of true-score theory. • Sources of Variation are viewed from different perspectives. • S2O = S2LG + S2TG + S2LS + S2TS L=long term, T=temporary G=general, S=specific
So what does this have to do with Multivariate Analysis? All techniques we have studied deals with the degree of overlap of variance (information) for sets of variables (Tabachnick & Fidell, 2001). In the present case the overlap of interest is between observed variance for evaluators’ opinions about the spheres involved and the actual underlying dimensions influencing interpersonal interactions perceived by the evaluators--“true-score” variance. S2TOT = S2E + S2V + S2Sc + S2e E=evaluator, V=vignette,Sc=scale, e=error
Relationship of Gerneralizability to Multivariance To produce the variance estimates needed ANOVA is used. In this situation a 6 x 7 x 46 Evaluator (E) x Sphere + Primary Category (S) x Vignette (V) Mixed Effects Repeated Measures ANOVA is used. Evaluator and Vignette are considered random, Sphere fixed.
ANOVA DESIGNDoubly Repeated MeasuresScale and Vignette WithinEvaluator Between
Table 1 Variance Components Derived from the SPSS GLM repeated Measures Calculation for Overall Evaluations Source of VariationSS dfMS Vignette (V) 29.220 45 .649 Sphere (S) 639.865 6 106.644 Evaluator (E) 61.972 5 12.394 V x S 185.516 270 .687 V x E 96.528 225 .429 S x E 340.184 30 11.339 V x S x E Error (e) Residual 648.149 1350 .480
Table 2 • Variance Components and Expected Mean Squares for Overall Evaluations • Source of VariationMeans Square (MS)Expected Mean Square (EMS)Estimated Variance • Vignette (V) .649 2e + + 2VE + 2V .220 • Sphere (S) 106.644 2e + + 2VS + 2SE + 2S ≥94.618 • Evaluator (E) 12.394 2e + + 2VE + 2E 11.965 • V x S .687 2e + + 2VS ≤.207 • V x E .429 2e + + 2VE ≤-.059 • S x E 11.339 2e + + 2SE ≤10.859 • V x S x E 2e + 2VSE ≥.480 • Error (e) 2e ≥.480 • Residual .480 .480 • Total (TOT) 2e + 2S + 2V + 2E ≤107.763 • ≥107.283 • = S2S / S2TOT ≥ 94.618 / 107.763 = 0.879 0.879 ≤ ≤ 0.886 ≤ 95.094 / 107.283 = 0.886
Table 3 Variance Components Derived from the SPSS GLM repeated Measures Calculation for Spheres Source of VariationSS dfMS Vignette (V) 12.426 45 .276 Sphere (S) 22.575 5 4.515 Evaluator (E) 16.996 5 3.399 V x S 73.730 225 .328 V x E 27.643 225 .123 S x E 24.769 25 .991 V x S x E Error (e) Residual 168.092 1125 .149
Table 4 • Variance Components and Expected Mean Squares for Spheres • Source of VariationMeans Square (MS)Expected Mean Square (EMS)Estimated Variance • Vignette (V) .276 2e + + 2VE + 2V .153 • Sphere (S) 4.515 2e + + 2VS + 2SE + 2S ≥3.345 • Evaluator (E) 3.399 2e + + 2VE + 2E 3.276 • V x S .328 2e + + 2VS ≤.179 • V x E .123 2e + + 2VE ≤-.026 • S x E .991 2e + + 2SE ≤.842 • V x S x E 2e + 2VSE ≥.149 • Error (e) 2e ≥.149 • Residual .149 .149 • Total (TOT) 2e + 2S + 2V + 2E ≤6.923 • ≥7.072 • = S2S / S2TOT ≥ 3.345 / 7.072 = 0.473 0.473 ≤ ≤ 0.505 ≤ 3.494 / 6.923 = 0.505
A Sample Calculation As an interesting example of how the ’s are calculated here is one (see Table 2 for the data) Sphere (MS) = 106.644 2e + + 2VS + 2SE + 2S -V x S (MS) = .687 2e + + 2VS -S x E (MS) = 11.339 2e + + 2SE +Error (MS) ≤ .480 2e Sphere (Variance) ≤ 95.094 2S -Error (MS) ≤ .480 2e Sphere (Variance) ≥ 94.618 Total (TOT) ≤ 107.763 2e + 2S + 2V + 2E ≥ 107.283 = S2S / S2TOT ≥ 94.618 / 107.763 = 0.879 ≤ 95.094 / 107.283 = 0.886 0.879 ≤ ≤ 0.882
The Conclusion What we get is three little numbers that are not as simple as they look: • Primary Influence: = .966 • Spheres: = .473 • Overall: = .879