prevalence, PPV and NPV

Comments

• EJ

  January 2022

  Hi Mani,

  Thanks for your question. Check out the Learn Bayes module and go to "Binary Classification"!

  Post: https://jasp-stats.org/2021/09/28/binary-classification/

  Cheers,

  E.J.
• mani

  January 2022

  Thankyou so much , I truly appreciate your guidance

  I don't see the Classical binary classification under any tab

  

  
  

  

  when I try Bayesian correlation it gives me no values

  Thankyou , please advise
• EJ

  January 2022

  1. Binary classification: Did you run the latest version? (it was recently added)
  2. Correlation: you need to enter a sample size (> 0)

  E.J.
• mani

  January 2022

  Hi

  now regardless of data I keep getting same values

  I ran 3 separate csv files with separate datasets from my machine

  but its giving me same results for all 3 which is not possible

  am I doing something wrong

  

  

  Thank you
• EJ

  January 2022

  You haven't loaded the data it seems.

  Cheers,

  E.J.
• mani

  January 2022

  whose data is it then ? I uploaded .csv file doing "point estimates" as input types, so are those default vales for point estimates

  

  are these default values for sensitivity (0.800) , prevalence (0.100) , specificity (0.800) TP , FP

  under Load data there are 2 category marker and positive condition

  time in months is my dependent variables and biomarkers phenotypes are independent variables , age- gender are covariates , Can I do this analysis do get PPV ,NPV I have 32 set of phenotypes

  also are alpha(8) beta(2) values are set by default

  it doesn't allow me to add more than 1 marker at a time and it doesn't allow me to add positive condition at all.

  Sorry for so many questions , I am new to this

  Thank you I truly appreciate your guidance

  

  
  
• Kucharssim

  January 2022

  Dear @mani,

  I would like to ask more details about your use case. It seems to me you have multiple predictors, in that case, the binary classification alone will not help you there. I have some scenarios in mind, please let me know if I am misunderstanding:

  1) If you have a binary dependent variable and multiple predictors, you can use logistic regression in the regression module. There, you can get performance metrics like sensitivity, specificity, accuracy, etc. Unfortunately the logistic regression does not compute PPV and NPV. If you think having PPV and NPV in logistic regression is exactly what you needed, we could add that as a feature request at: https://github.com/jasp-stats/jasp-issues so that we can implement it in JASP. A quick and dirty alternative to compute the PPV and NPV would be by combining the output of the logistic regression and the binary classification:

  a) carry out the logistic regression with the independent variables as you wanted. Then, select "confusion matrix", "specificity", and "sensitivity". Then, open the binary classification module, select "point estimates". For prevalence, compute the proportion of cases that are in the confusion matrix in the observed positive category. Fill in the sensitivity and specificity in the other two fields. This will give you access to the PPV and NPV from that analysis, see screenshot:

  

  b) in case you want to estimate the uncertainty in PPV and NPV, you can carry out the logistic regression, select "confusion matrix", and then open binary classification and select "uncertain estimates". Then, fill in the counts from the confusion matrix as observed data for the binary classification. This will give you also credible intervals, etc., see screenshot

  

  note that in the latter case, your results are affected by the priors you specify in the analysis - the point estimates from "uncertain estimates" will not correspond to point estimates from the "point estimates" method.

  2) I see that you write "time in months" is your dependent variable. Could you elaborate what do you want to do exactly? I can't wrap my head around in what case could we use PPV and NPV when we are predicting a continuous variable. Maybe it's a misunderstanding, but at the moment, I am unfortunately not sure how to help you with that... If you elaborate, perhaps we can get to a solution!

  
  

  Best,

  Simon
• mani

  January 2022

  Thankyou so much Simon,

  I truly appreciate your guidance :)

  I have for each study ID

  Time in months is – dependent variable

  Biomarkers (1 to 32) –

  Covariates – age, gender

  Trying to see if there is any correlation causation between any of the biomarkers to survival in months (if the biomarkers can extend the life of any study id)

  & there are 3 grades of tumor from grade 2 to grade 3 and 3 panels each grade

  Grade 2- panel 1,2,3

  Grade 3- panel 1,2,3

  Grade 4 – panel 1,2,3

  Is it possible to do

  1.    Cross Correlation- between all 32 markers of all 3 panels

  2.    Self-correlation – grade 2 v/s grade 3 , grade 3 v/s grade 4 and grade 4 v/s grade 2

  3.    Univariable correlation – between 32 biomarkers

  This is truly a confusion matrix name is apt for it

  I have 32 rows and 32 columns in Pearson’s heatmap but while exporting results as PDF it cuts off at column 15 on right and remaining columns are not exported

  The only way it allows me to save results is JASP file which opens in software but how can I export it as a whole in one frame

  Thankyou so much