Timeline for Correlation between Likert-type item and yes / no question
Current License: CC BY-SA 3.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 2, 2013 at 17:31 | comment | added | Peter Flom | @ttnphns No, it's not always arbitrary, and there can be additional evidence. Some scales are interval by their nature (e.g temp. in degrees C) but even some likert wordings have some evidence (although it's been a long time since I looked at it) | |
| Jul 2, 2013 at 13:09 | comment | added | ttnphns |
@Peter, Again: it's not the regression I was against, it's condemn to "Pearson or Kendall". Your last comment makes any more or less sense actually works contra that sentence. The decision to treat it as ordinal or (equi)interval is ultimately untestible, it is always an arbitrary assumption; no additional evidence is needed to appoint a scale to be interval instead of ordinal.
|
|
| Jul 2, 2013 at 11:45 | comment | added | Peter Flom | @ttnphns You can treat a 3 point scale as interval, but the assumption behind that treatment seem unreasonable and untestable. There is no reason that I can see that (say) agree = 2, disagree = 1, strongly disagree = 0 makes any more or less sense than (say) agree = 5, disagree = -5, strongly disagree = -10. Given that, you could do point-biserial rank correlation, but what would that tell you? I still think the question calls for regression. | |
| Jul 2, 2013 at 8:40 | comment | added | ttnphns | (Cont.) I then concluded that Peter's claim "Neither Pearson nor Kendall correlation really seems suited here" is untrue. These coefficients may be used. Because one of the two variables is that "unique" and the other could formally be assumed either ordinal of interval. And I never said logistic regression is not reasonable here. | |
| Jul 2, 2013 at 8:35 | comment | added | ttnphns | @gung, I agree practically and disagree theoretically. Formally, any quantitative scale with 3+ points may be taken as interval scale (as well as ordinal scale); dilemma "interval or ordinal?" is logically independent of the number of points which is the question of descretness, the degree of bluntness of measuring capacity. As for 2-points scale, it is unique in that respect that since it has just 1 interval we can't judge if it is ordinal or interval. I said that in many cases it behaves statistically identically - whether treated as "interval", "ordinal" or "nominal". | |
| Jul 2, 2013 at 2:07 | comment | added | gung - Reinstate Monica | @ttnphns, the job satisfaction variable is a single Likert item w/ only 3 levels. I fail to see how treating that as interval data is the best way to go. Logistic regression seems reasonable. | |
| Jul 1, 2013 at 23:32 | comment | added | Elle | Well, thank you for your answer. Actually, my independent variable is job satisfaction and my dependent variable is intention to stay. | |
| Jul 1, 2013 at 23:18 | comment | added | ttnphns | The Likert-type rating scale could be assumed to be ordinal or inteval. The dichotomous question is formally nominal, but since it is of 2 categories it is statistically equivalent to interval or ordinal (proof 1: if both variables are dichotomous, Phi = Pearson r = Kendall tau-b = Spearman rho = Eta; proof 2: dichotomous IV behaves identically as factor and as [when coded binary] covariate). So, your statement "neither Pearson nor Kendall suits here" is wrong. | |
| Jul 1, 2013 at 22:30 | comment | added | Peter Flom | Because 1) Neither variable is numeric; point biserial would work if one was numeric and one was binary. 2) Regression seems to be what is needed, as there is a clear DV. | |
| Jul 1, 2013 at 21:48 | comment | added | ttnphns |
Neither Pearson nor Kendall correlation really seems suited here Why so, Peter? Would you support your claim?
|
|
| Jul 1, 2013 at 21:29 | history | edited | Nick Cox | CC BY-SA 3.0 |
edited body
|
| Jul 1, 2013 at 21:19 | history | answered | Peter Flom | CC BY-SA 3.0 |