Yes, ordered logit or probit would be where to start. Here's a tutorialtutorial on ordered logit that uses R. Other CV questions can probably help you with any snags you run into—try the tags 'logit,' 'probit,' and 'ordinal.'
A standard approach to dealing with a categorical independent variable with $k$ values is to dummy code it as $k-1$ binary values. This is more fully explained herehere, but in short: The effect of one category is subsumed into the intercept, and coefficients are fitted to the remaining categories. In your example, there would be a dummy variable
caucasianthat would be coded to 1 for a Caucasian respondent, 0 otherwise.Dealing with missing data very much depends on the problem at hand, and yes, how you deal with missing data may introduce bias. This book excerpt nicely describes four mechanisms that can produce missing data, which should help you consider potential bias in your own problem at hand. (In particular, section 25.1, p. 530.)
Many modeling packages have a
predictfunction of some sort, and indeed the first tutorial linked above includes a demonstration.
Yes, ordered logit or probit would be where to start. Here's a tutorial on ordered logit that uses R. Other CV questions can probably help you with any snags you run into—try the tags 'logit,' 'probit,' and 'ordinal.'
A standard approach to dealing with a categorical independent variable with $k$ values is to dummy code it as $k-1$ binary values. This is more fully explained here, but in short: The effect of one category is subsumed into the intercept, and coefficients are fitted to the remaining categories. In your example, there would be a dummy variable
caucasianthat would be coded to 1 for a Caucasian respondent, 0 otherwise.Dealing with missing data very much depends on the problem at hand, and yes, how you deal with missing data may introduce bias. This book excerpt nicely describes four mechanisms that can produce missing data, which should help you consider potential bias in your own problem at hand. (In particular, section 25.1, p. 530.)
Many modeling packages have a
predictfunction of some sort, and indeed the first tutorial linked above includes a demonstration.
Yes, ordered logit or probit would be where to start. Here's a tutorial on ordered logit that uses R. Other CV questions can probably help you with any snags you run into—try the tags 'logit,' 'probit,' and 'ordinal.'
A standard approach to dealing with a categorical independent variable with $k$ values is to dummy code it as $k-1$ binary values. This is more fully explained here, but in short: The effect of one category is subsumed into the intercept, and coefficients are fitted to the remaining categories. In your example, there would be a dummy variable
caucasianthat would be coded to 1 for a Caucasian respondent, 0 otherwise.Dealing with missing data very much depends on the problem at hand, and yes, how you deal with missing data may introduce bias. This book excerpt nicely describes four mechanisms that can produce missing data, which should help you consider potential bias in your own problem at hand. (In particular, section 25.1, p. 530.)
Many modeling packages have a
predictfunction of some sort, and indeed the first tutorial linked above includes a demonstration.
Yes, ordered logit or probit would be where to start. Here's a tutorial on ordered logit that uses R. Other CV questions can probably help you with any snags you run into—try the tags 'logit,' 'probit,' and 'ordinal.'
A standard approach to dealing with a categorical independent variable with $k$ values is to dummy code it as $k-1$ binary values. This is more fully explained here, but in short: The effect of one category is subsumed into the intercept, and coefficients are fitted to the remaining categories. In your example, there would be a dummy variable
caucasianthat would be coded to 1 for a Caucasian respondent, 0 otherwise.Dealing with missing data very much depends on the problem at hand, and yes, how you deal with missing data may introduce bias. This book excerpt nicely describes four mechanisms that can produce missing data, which should help you consider potential bias in your own problem at hand. (In particular, section 25.1, p. 530.)
Many modeling packages have a
predictfunction of some sort, and indeed the first tutorial linked above includes a demonstration.