# What analysis should I run?

#### audreye

##### New Member
Hi,

I am not sure which analysis to run based on what I am trying to figure out, any guidance is appreciated.

The research question is- are certain ethnicities more likely to report their ethnicity in the online classroom?

Data- I have a variable of ethnicities (1= Caucasian, 2= African American, etc.). I have another variable in which participants used a likert scale (1= agree, 2= disagree, etc.) to a question regarding their comfort level in reporting ethnicity in the classroom.

My question is, can I run an analysis that compares the two variables and gives me frequencies for each individual ethnicity. For example, how frequent is agree among Caucasian, among African American, etc. and how frequent is disagree among Caucasian....

I initially thought a correlation would work but that doesn't give me the individual frequencies for each ethnic group.

Thank you!!!

#### trinker

##### ggplot2orBust
audreye said:
I have another variable in which participants used a likert scale (1= agree, 2= disagree, etc.) to a question regarding their comfort level in reporting ethnicity in the classroom.
This is not a scale but an item with responses. A Likert scale is multiple questions. With what you've given us I think you have a categorical IV and a categorical DV. If missing data isn't a problem (and empty cells; i.e., check assumptions) a chi square may be appropriate here.

#### CB

##### Super Moderator
Categorical/nominal IV, ordered categorical (i.e. ordinal) DV > Kruskal Wallis test?

#### brandt

##### New Member
Nominal logistic regression perhaps?

#### noetsi

##### No cake for spunky
Well if the DV is ordered I would think that ordered logistic regression makes more sense, with dummy IV for the categorical variables. If the assumptions of ordered logistic regression are met (it has one special one beyond that of logistic regression only which has a specific statistical test).

This is not a scale but an item with responses. A Likert scale is multiple questions.
I understand this is technically true, but it is extremely common to talk of an ordered question (that is when responses are ordinal in nature) as "likert scale." You could probably find a thousand articles with that usage.

#### CB

##### Super Moderator
I understand this is technically true, but it is extremely common to talk of an ordered question (that is when responses are ordinal in nature) as "likert scale." You could probably find a thousand articles with that usage.
Maybe, but right from the first paragraph of the WP article:
Likert distinguished between a scale proper, which emerges from collective responses to a set of items (usually eight or more), and the format in which responses are scored along a range.
The difference between a scale and an item does matter for someone trying to select which analysis to use. E.g., in this case if the DV was actually a Likert scale, with many possible values, then your otherwise-sensible suggestion of ordinal logistic regression might not work so neatly, because there'd be a very large number of threshold parameters to estimate. So being finicky about the difference can actually be helpful.

#### noetsi

##### No cake for spunky
If they meant a series of likert questions collapsed into a single multipoint scale, I was told in my SEM courses and read elsewhere, they can normally treat that as interval like and thus use linear regression. In fact it is common in SEM to collapse multiple likert questions into one response and use them as interval.

I should just have ignored commenting seperately on the previous statement about likert scale. I did so because this is a common issues, using likert scale although what is really meant is a likert like question (that is a single question with ordered values).