Good content. I have a question : 1.) What will be the structure if labelled dataset to train for hazard ? Can we expect any business setting will have hazard function labelled in the dataset ? that looks quite impossible. How then in that case, it can be used for training(estimating the b0,b1,...) etc. ?
If you want to estimate the hazard, Exponential or Weibull are good options. But note they assume constant hazard and a hazard that changes proportionally with time, respectively
Thanks a lot for the great content. I have one question regarding constancy of the hazard. As mentioned in the lecture, lambda is replaced by hazard and lambda is constant. Doesn't it follow that hazard is constant as well? If that's true, that means there is no reason to do regression, as the dependent variable of the regression equation is constant. Could you please explain how that part work?
Good question. What is meant is that the hazard is constant over time, but it varies depending on the values of the X variables. (Eg) the hazard will be different for a male and female, or different depending on age, and so forth. BUT, consider a female of a given age...their hazard will be the same at any given point in time. So, we use a regression model to estimate the hazard for a given set of X variables, but assume that their hazard does not change over time
One more question: and the hazard depending on X variables will be calculated using "small" population, where for each person the dependent variables are the same. Am I true?
Yes, that’s right. For example, suppose you had only X1=sex and X2=Age, then for the model you would be estimating the hazard for a given sex and age. So, if you input sex of male and age 30, then the model would return the estimated hazard (which would give you the estimated survival function) for males of age 30...and it is assumed that this is the same for all males of age 30. Of course, we know that it won’t be exactly the same for all males of age 30, but with no other input variables this is all we are estimating. You can think of it as sort of being the average hazard for males of age 30. This is in fact how all these regression models (like linear or logistic) work. They estimate the “average” Y for a given set of X’s
can I get a great link for Poisson regression basics. Also, my understanding is that (based on so far and not any further video) in Poisson regression, we have held 't' as constant hence it is missing in the equation. However, it would be wrong to assume that this 't' is constant in a cohort study because not everyone has same followup time. And therefore, when dealing with Poisson, in cohort setting, we do have the time (person-time) in the equation. is this correct?
I have a bunch of videos for poisson regression. Start with video 9.1 (linked to here ruclips.net/video/9YrobQt5Hs8/видео.html ) and watch all the 9.x and 10.x videos for a full explanation of poisson regression. You are correcting that in poisson the t is fixed and we ask about x, the number of occurrences in time t. The model can account for people having different follow up times (varying value if t for individuals) by using an offset. How this is done is explained in the videos. These videos cover 2 weeks of a course I teach , with weeks 9 and 10 discussing poisson regression. (And weeks 11-12 covering survival analysis )
We had forgotten to upload that one..should add it soon. It is just another example and a bit repetitive…it is another example of the material from the first 5 videos
wait I got it. the definition of hazard P(Tt ) means that hazard at time t is the death number right at time t divided by the number at risk a little bit before time t. So, here we have the death number at time t, which is -S'(t) (note that the derivative is negative but death number is positive, so we need to add a minus before), and number at risk S(t), then we have hazard = -S'(t)/S(t) = lambda
best content ever seen on RUclips about SA, thank u and wish u best, keep up ur good work
Very good series, but the part 6 seems to be missing?
Did you find the Part 6?
@@mahery_ranaivoson Nope
Thank you. I am very much grateful.
Fantastic series, thank you very much
I am dying to know what happened to part 6 😀😄
Excellent lecture!
Great video. part 6 is missing, or maybe a typo in the titles? great work nonetheless
Very nicely explained... Keep it up... Best of luck...
Very well explained!
Could you explain the remark about exp(b) = hazard ratio (14:25) please? Was it explained in the nowhere to be found) part 6?
Zhiguo Wang Thanks for the explanation. So was there a part 6?
I think part 6 has been deleted? Great series nonetheless :)
Good content. I have a question : 1.) What will be the structure if labelled dataset to train for hazard ? Can we expect any business setting will have hazard function labelled in the dataset ? that looks quite impossible. How then in that case, it can be used for training(estimating the b0,b1,...) etc. ?
Thanks for the video! One question: why do you use Poisson (instead of linear) regression to estimate Hazard (around 14:00)?
Because poisson regression is for estimating a rate, and the hazard is the rate for survival function
Thanks for sharing
you are amaaaaaaazing , thank you very much , may god please you
Hi, thank you for the video! Which regression model do you use to get the numbers to estimate the hazard?
If you want to estimate the hazard, Exponential or Weibull are good options. But note they assume constant hazard and a hazard that changes proportionally with time, respectively
Anyone wondering if for the poisson law its the Big Y or small y on the denominator. I've checked it should be the small y. At minute 2:32
Hi! Could you please tell me what is b0,b1 ... here is HAZ? x1,x2... these are variables like age, height etc. Right?
Thanks a lot for the great content. I have one question regarding constancy of the hazard. As mentioned in the lecture, lambda is replaced by hazard and lambda is constant. Doesn't it follow that hazard is constant as well? If that's true, that means there is no reason to do regression, as the dependent variable of the regression equation is constant. Could you please explain how that part work?
Good question. What is meant is that the hazard is constant over time, but it varies depending on the values of the X variables. (Eg) the hazard will be different for a male and female, or different depending on age, and so forth. BUT, consider a female of a given age...their hazard will be the same at any given point in time. So, we use a regression model to estimate the hazard for a given set of X variables, but assume that their hazard does not change over time
Thanks a lot. Clear!!!
One more question: and the hazard depending on X variables will be calculated using "small" population, where for each person the dependent variables are the same. Am I true?
Yes, that’s right. For example, suppose you had only X1=sex and X2=Age, then for the model you would be estimating the hazard for a given sex and age. So, if you input sex of male and age 30, then the model would return the estimated hazard (which would give you the estimated survival function) for males of age 30...and it is assumed that this is the same for all males of age 30. Of course, we know that it won’t be exactly the same for all males of age 30, but with no other input variables this is all we are estimating. You can think of it as sort of being the average hazard for males of age 30.
This is in fact how all these regression models (like linear or logistic) work. They estimate the “average” Y for a given set of X’s
Thanks a lot.
Hey, is there a part 6?
can I get a great link for Poisson regression basics. Also, my understanding is that (based on so far and not any further video) in Poisson regression, we have held 't' as constant hence it is missing in the equation. However, it would be wrong to assume that this 't' is constant in a cohort study because not everyone has same followup time. And therefore, when dealing with Poisson, in cohort setting, we do have the time (person-time) in the equation. is this correct?
I have a bunch of videos for poisson regression. Start with video 9.1 (linked to here ruclips.net/video/9YrobQt5Hs8/видео.html ) and watch all the 9.x and 10.x videos for a full explanation of poisson regression.
You are correcting that in poisson the t is fixed and we ask about x, the number of occurrences in time t. The model can account for people having different follow up times (varying value if t for individuals) by using an offset. How this is done is explained in the videos.
These videos cover 2 weeks of a course I teach , with weeks 9 and 10 discussing poisson regression. (And weeks 11-12 covering survival analysis )
Thanks for your video but I'm wondering where is part 6? 😅
We had forgotten to upload that one..should add it soon. It is just another example and a bit repetitive…it is another example of the material from the first 5 videos
The lambda can be larger than 1. But the hazard is the probability(always < 1) of a events occurs in next period of time. So why lambda = haz here?
Same question, in part 2 we define hazard as P(Tt ), I try to calculate this using conditional probability but fail to prove that lambda is hazard
wait I got it. the definition of hazard P(Tt ) means that hazard at time t is the death number right at time t divided by the number at risk a little bit before time t. So, here we have the death number at time t, which is -S'(t) (note that the derivative is negative but death number is positive, so we need to add a minus before), and number at risk S(t), then we have hazard = -S'(t)/S(t) = lambda