Timeline for SPSS - Binary logistic regression: classification cutoff
Current License: CC BY-SA 4.0
17 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 20, 2018 at 23:11 | vote | accept | FSJ963 | ||
| S Oct 20, 2018 at 23:09 | history | bounty ended | FSJ963 | ||
| S Oct 20, 2018 at 23:09 | history | notice removed | FSJ963 | ||
| Oct 19, 2018 at 18:00 | history | tweeted | twitter.com/StackStats/status/1053345159457452032 | ||
| Oct 18, 2018 at 23:44 | answer | added | EdM | timeline score: 5 | |
| S Oct 15, 2018 at 21:25 | history | bounty started | FSJ963 | ||
| S Oct 15, 2018 at 21:25 | history | notice added | FSJ963 | Authoritative reference needed | |
| Oct 14, 2018 at 9:09 | comment | added | FSJ963 | And hence, influencing the pretest probability of a person belonging to one of the two groups. | |
| Oct 14, 2018 at 8:10 | comment | added | FSJ963 | Taking into account a difference in prevalence between 2 groups in a population in the prediction of a person belonging to one of the 2 groups with binary logistic regression | |
| Oct 14, 2018 at 2:08 | comment | added | user158565 | What is your purpose of setting up a cutoff point? | |
| Oct 13, 2018 at 23:16 | comment | added | FSJ963 | Sorry for the messy comment, but I wanted to show you my line of thought, right after I recalled that I have to calculate with odds rather than probabilities when taking into account arguments etc. PrTP = pre test probability; PrTO = pre test odds; ARG = argument; MAL = malignant tumour; CUTO = cutoff odds; PoTO = post test odds. So my question in short: can I set the classification cutoff value to 0.30? | |
| Oct 13, 2018 at 23:14 | comment | added | FSJ963 | SPSS: "PrTP = 0.50" or "PrTO = 1.00" || PrTO * ARG = PoTO || SPSS: 1.00 * ARG = PoTO || SPSS: if 1.00 * ARG > 1.00 then MAL || 2 options to incorporate true pretest probability: (1) Increase PrTP or PrTO such that PrTO * ARG > CUTO (impossible in SPSS) (2) Decrease CUTO such that PrTO * ARG > CUTO || REAL: PrTP = 0.70 | PrTO = 2.33 || REAL: 2.33 * ARG = PoTO || REAL: if 2.33 * ARG > 1.00 then MAL || ((1) Increase PrTO: 2.33 * ARG > 1.00) || (2) Decrease CUTO: 1.00 * ARG > 0.43 NEW CUTO: 0.43 ~ P = 30 % | |
| Oct 13, 2018 at 22:38 | comment | added | FSJ963 | ... here the cutoff of the probability cannot be 50 %, since there are more malignant tumours in the population than benign tumours. And although that is not the case in my sample, we need to account for that fact. We need to incorporate a 70 % a priori probability for malignancy, rather than 50 %; and since this is not directly possible in SPSS, I was wondering whether I could instead change the malignancy probability cutoff: | |
| Oct 13, 2018 at 22:34 | comment | added | FSJ963 | @a_statistician the problem with that is that that would be based on my sample, which consists of only about 30 patients. In my sample, there's about an even number of persons with a malignant tumour and persons with a benign tumour. We're not talking about a cutoff value of the variable, but in logistic regression, the cutoff point of the probability for malignancy: a tumour has a probability for malignancy ranging from 0 to 100 %, and the cutoff is somewhere in between: when there is an even amount of malignant and benign tumours in the population, that would be 50 %. But ... | |
| Oct 13, 2018 at 21:31 | comment | added | kjetil b halvorsen♦ | Maybe a duplicate: stats.stackexchange.com/questions/212228/… Also stats.stackexchange.com/questions/67091/… | |
| Oct 13, 2018 at 21:13 | comment | added | user158565 | Check the sensitivity and specificity at the different cut-off point (between 0 to 1). ROC curve is helpful. Then find the best cut-off point. | |
| Oct 13, 2018 at 20:16 | history | asked | FSJ963 | CC BY-SA 4.0 |