Timeline for Correlating perception with an objective measure - average ratings vs replicating the objective measure
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Sep 18, 2019 at 11:18 | comment | added | user2974951 | @brainfullofdoubt Yes of course, however you did not obtain such multiple values of y (for a point in x), you merely produced duplicates out of thin air, which will also decrease the variance of y. Technically speaking, if y is truly a continuous variable, then the probability of obtaining the same value even twice is 0. | |
| Sep 18, 2019 at 3:30 | vote | accept | brainfullofdoubt | ||
| Sep 18, 2019 at 3:30 | comment | added | brainfullofdoubt | Thank you. So I’ll mark it as answered now. However I do not fully understand your last comment. It may still be good replying, for anyone else who reads this later, as well as for me! ....... Generally regression predicts a mean, which might be written y = ax + b + noise. This implies that there is a variance about that mean, which also leads that there can be multiple “y” values at a single “x”, no need to replicate x. | |
| Sep 17, 2019 at 8:50 | comment | added | user2974951 | @brainfullofdoubt This would require you to replicate your true scores. You would be making up new unobserved values which do not exist. And since you are just replicating them, you would decrease the uncertainty in the DV. In other words, you would be analyzing the effect that your IV have on a fictitious value, that these IV are supposed to produce. Since these two scores are obtained from two different sources, one does not cause the other, also the true scores are fixed, there is no error in them. | |
| Sep 16, 2019 at 12:40 | comment | added | brainfullofdoubt | One more question, and then I will mark as answered. If the objective is true, why not then just handle this as a regression where the objective is the independent variable, and user scores are the dependent. The regression would then give r^2 and the amount of the overall variance explained by the independent variable (I believe)? | |
| Sep 16, 2019 at 12:39 | comment | added | brainfullofdoubt | Thank you for the reply. Assuming the objective value is "true" feels unnatural, since we are trying to validate if the objective measure is a good surrogate for the human. However, I'll go with that... | |
| Sep 15, 2019 at 19:39 | comment | added | user2974951 | @brainfullofdoubt Well, you sure took your time. See my edits. | |
| Sep 15, 2019 at 19:39 | history | edited | user2974951 | CC BY-SA 4.0 |
added 2004 characters in body
|
| Sep 13, 2019 at 13:55 | comment | added | brainfullofdoubt | But more important, why is code box 1 the right thing to do. It seems that it is replicating the objective measure. We need to be able to defend and explain the chosen method. Thank you. Also big sorry that we did not see your reply. It after a couple days assumed that no one could answer. | |
| Sep 13, 2019 at 13:55 | comment | added | brainfullofdoubt | This looks like a careful answer. However can you explain (for beginners) more of why this is a good answer to the question? In other words, keep the code, but add some more explaination saying why you picked this approach. For example, in the second part, why does examining within-object differences relate to how well the objective measure relates to user rating. I think the numbers (1,0,0) etc in the variable K may have typos, or else I do not understand what is meant. | |
| Feb 7, 2019 at 9:06 | history | answered | user2974951 | CC BY-SA 4.0 |