13.11.2012 Views

Introduction to Categorical Data Analysis

Introduction to Categorical Data Analysis

Introduction to Categorical Data Analysis

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

8.5 ANALYZING RATER AGREEMENT 261<br />

that they both classify a subject in category i. The sum �<br />

i πii is the <strong>to</strong>tal probability<br />

of agreement. Perfect agreement occurs when �<br />

i πii = 1.<br />

Many categorical scales are quite subjective, and perfect agreement is rare. This<br />

section presents ways <strong>to</strong> measure strength of agreement and detect patterns of<br />

disagreement. Agreement is distinct from association. Strong agreement requires<br />

strong association, but strong association can exist without strong agreement. If<br />

observer X consistently classifies subjects one level higher than observer Y , the<br />

strength of agreement is poor even though the association is strong.<br />

8.5.1 Cell Residuals for Independence Model<br />

One way of evaluating agreement compares the cell counts {nij } <strong>to</strong> the values<br />

{ni+n+j /n} predicted by the loglinear model of independence (7.1). That model provides<br />

a baseline, showing the degree of agreement expected if no association existed<br />

between the ratings. Normally it would fit poorly if there is even only mild agreement,<br />

but its cell standardized residuals (Section 2.4.5) provide information about patterns<br />

of agreement and disagreement.<br />

Cells with positive standardized residuals have higher frequencies than expected<br />

under independence. Ideally, large positive standardized residuals occur on the main<br />

diagonal and large negative standardized residuals occur off that diagonal. The sizes<br />

are influenced, however, by the sample size n, larger values tending <strong>to</strong> occur as n<br />

increases.<br />

In fact, the independence model fits Table 8.7 poorly (G2 = 118.0,df = 9).<br />

Table 8.7 reports the standardized residuals in parentheses. The large positive standardized<br />

residuals on the main diagonal indicate that agreement for each category<br />

is greater than expected by chance, especially for the first category. The off-maindiagonal<br />

residuals are primarily negative. Disagreements occurred less than expected<br />

under independence, although the evidence of this is weaker for categories closer<br />

<strong>to</strong>gether. Inspection of cell counts reveals that the most common disagreements refer<br />

<strong>to</strong> observer Y choosing category 3 and observer X instead choosing category 2 or 4.<br />

8.5.2 Quasi-Independence Model<br />

A more useful loglinear model adds a term that describes agreement beyond that<br />

expected under independence. This quasi-independence model is<br />

log μij = λ + λ X i + λY j + δiI(i = j) (8.12)<br />

where the indica<strong>to</strong>r I(i = j) equals 1 when i = j and equals 0 when i �= j. This<br />

model adds <strong>to</strong> the independence model a parameter δ1 for cell (1, 1) (in row 1 and<br />

column 1), a parameter δ2 for cell (2, 2), and so forth. When δi > 0, more agreements<br />

regarding outcome i occur than would be expected under independence. Because of

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!