Method Framework
Consumer behavior and potential influencing factors in this study are measured with a
mix of binary, ordinal, nominal, and cardinal data. Most response variables of interests
are non-cardinal. Hence, models for categorical response variables are chosen as the
major tools of analysis. Categorical analysis methods have been used in many fields of
social investigation, especially when data are gathered through survey (Agresti; Greene).
The model-building paradigm, including the linear probability models (LPM), probit
models, and logit models, is more informative than others for its focus on estimating
parameters and assessing effects of factors related. Varying in some degree, the three
models share more similarities. A common for all models is categorical response variable,
which could be the binary response (eating or not eating goat meat), the ordinal response
(ratings of an attribute: important, neutral, and unimportant), or the nominal response
(seasonal consumption preferences: winter, summer, fall or spring). For binary and
multinomial response variables, logistic models with general logit functions are usually
used. For ordinal response variables, ordinal logistic models with accumulative logit
functions are suitable.
In the binary case, we have a response yi for observation unit i. yi equals 1 if the
event of interest occurs for the ith observation unit, equals 0 if the event of interest does
not occur for the ith observation unit. The density function of yi is where Piis E (yi),
(1) f(yt) = -J!—7P (1 -P)yi) yi = 0,1
yi!*(1-yi)