search

rpietro / EsophagealCancerT1N0

Evaluating outcomes after Esophageal cancer T1N0
0 stars 0 forks source link

Comparisons across different datasets #6

Closed mworni closed 12 years ago

mworni commented 12 years ago

I created now a nsqip.sample and a nsqip.sample_IMPUTED dataset. How can I compare bmi in the first with the latter dataset?

variable name is bmi in both datasets. bmi__imp is a variable that indicates if the variable was imputed or not (coded 0 and 1).

rpietro commented 12 years ago

i'm not sure i understand the question, but:

  1. graphically through the VIM functions to check where the location is
  2. through summary measures and inferential tests (t-tests) to see whether the two are different
  3. comparing models adjusted for one vs. the other

you already know all of this, which tells me that you are asking something different

On Sun, Jul 1, 2012 at 5:21 AM, mworni < reply@reply.github.com

wrote:

I created now a nsqip.sample and a nsqip.sample_IMPUTED dataset. How can I compare bmi in the first with the latter dataset?

variable name is bmi in both datasets. bmi__imp is a variable that indicates if the variable was imputed or not (coded 0 and 1).

Reply to this email directly or view it on GitHub: https://github.com/rpietro/EsophagealCancerT1N0/issues/6

mworni commented 12 years ago

the question is much more technical...

if you check the script - there is a ton of stuff new in relation to the imputation. It definitely works with the sample set - and I can make also a ton of comparisons between the original and the old dataset using VIM. Very nice graphs and all as code now in the script.

However, what if I would like to make a t-test between the imputed and non imputed bmi? How do I have to code this.

Can I run a t-test between the nsqip.sample and the nsqip.sample_IMPUTED dataset? Do I have to make a t-test between bmi original and bmi imputed using bmi__imp as an indicator variable? However, this would give me two different results!

Awesome stuff!!!!! Looks great!

On Sun, Jul 1, 2012 at 1:42 PM, Ricardo Pietrobon < reply@reply.github.com

wrote:

i'm not sure i understand the question, but:

  1. graphically through the VIM functions to check where the location is
  2. through summary measures and inferential tests (t-tests) to see whether the two are different
  3. comparing models adjusted for one vs. the other

you already know all of this, which tells me that you are asking something different

On Sun, Jul 1, 2012 at 5:21 AM, mworni < reply@reply.github.com

wrote:

I created now a nsqip.sample and a nsqip.sample_IMPUTED dataset. How can I compare bmi in the first with the latter dataset?

variable name is bmi in both datasets. bmi__imp is a variable that indicates if the variable was imputed or not (coded 0 and 1).

Reply to this email directly or view it on GitHub: https://github.com/rpietro/EsophagealCancerT1N0/issues/6

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6692811

rpietro commented 12 years ago

i'm still not entirely sure i understand the question, but:

  1. if you have two variables (bmi and bmi_imp) you should run a t.test using something like t.test(bmi, bmi_imp) , but i think you've done this before. is your question whether the test should be paired or unpaired? sorry, i still don't think i understand it
  2. if you are trying to compare the two datasets as a whole (all imputed variables compared at once) then you would have to compare beta coefficients and CIs, but i think you also know that

On Sun, Jul 1, 2012 at 7:48 AM, mworni < reply@reply.github.com

wrote:

the question is much more technical...

if you check the script - there is a ton of stuff new in relation to the imputation. It definitely works with the sample set - and I can make also a ton of comparisons between the original and the old dataset using VIM. Very nice graphs and all as code now in the script.

However, what if I would like to make a t-test between the imputed and non imputed bmi? How do I have to code this.

Can I run a t-test between the nsqip.sample and the nsqip.sample_IMPUTED dataset? Do I have to make a t-test between bmi original and bmi imputed using bmi__imp as an indicator variable? However, this would give me two different results!

Awesome stuff!!!!! Looks great!

On Sun, Jul 1, 2012 at 1:42 PM, Ricardo Pietrobon < reply@reply.github.com

wrote:

i'm not sure i understand the question, but:

  1. graphically through the VIM functions to check where the location is
  2. through summary measures and inferential tests (t-tests) to see whether the two are different
  3. comparing models adjusted for one vs. the other

you already know all of this, which tells me that you are asking something different

On Sun, Jul 1, 2012 at 5:21 AM, mworni < reply@reply.github.com

wrote:

I created now a nsqip.sample and a nsqip.sample_IMPUTED dataset. How can I compare bmi in the first with the latter dataset?

variable name is bmi in both datasets. bmi__imp is a variable that indicates if the variable was imputed or not (coded 0 and 1).

Reply to this email directly or view it on GitHub: https://github.com/rpietro/EsophagealCancerT1N0/issues/6

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6692811

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6692852

mworni commented 12 years ago

I just realized that this is under the wrong label - this question relates to NSQIP and not esophageal cancer.

It's hard to explain since it comes out of VIM. When you impute bmi using kNN, then you get for every patient a bmi value and an additional variable called bmi__imp which indicates for each individual if bmi was imputed or original.

I now wanted to compare bmi from the original dataset with the bmi values I got after the imputation. But maybe this is not possible.

On Sun, Jul 1, 2012 at 2:41 PM, Ricardo Pietrobon < reply@reply.github.com

wrote:

i'm still not entirely sure i understand the question, but:

  1. if you have two variables (bmi and bmi_imp) you should run a t.test using something like t.test(bmi, bmi_imp) , but i think you've done this before. is your question whether the test should be paired or unpaired? sorry, i still don't think i understand it
  2. if you are trying to compare the two datasets as a whole (all imputed variables compared at once) then you would have to compare beta coefficients and CIs, but i think you also know that

On Sun, Jul 1, 2012 at 7:48 AM, mworni < reply@reply.github.com

wrote:

the question is much more technical...

if you check the script - there is a ton of stuff new in relation to the imputation. It definitely works with the sample set - and I can make also a ton of comparisons between the original and the old dataset using VIM. Very nice graphs and all as code now in the script.

However, what if I would like to make a t-test between the imputed and non imputed bmi? How do I have to code this.

Can I run a t-test between the nsqip.sample and the nsqip.sample_IMPUTED dataset? Do I have to make a t-test between bmi original and bmi imputed using bmi__imp as an indicator variable? However, this would give me two different results!

Awesome stuff!!!!! Looks great!

On Sun, Jul 1, 2012 at 1:42 PM, Ricardo Pietrobon < reply@reply.github.com

wrote:

i'm not sure i understand the question, but:

  1. graphically through the VIM functions to check where the location is
  2. through summary measures and inferential tests (t-tests) to see whether the two are different
  3. comparing models adjusted for one vs. the other

you already know all of this, which tells me that you are asking something different

On Sun, Jul 1, 2012 at 5:21 AM, mworni < reply@reply.github.com

wrote:

I created now a nsqip.sample and a nsqip.sample_IMPUTED dataset. How can I compare bmi in the first with the latter dataset?

variable name is bmi in both datasets. bmi__imp is a variable that indicates if the variable was imputed or not (coded 0 and 1).

Reply to this email directly or view it on GitHub: https://github.com/rpietro/EsophagealCancerT1N0/issues/6

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6692811

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6692852

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6693149

rpietro commented 12 years ago

got it now. here is some mock code:

bmi_imp <- bmi

do your imputation stuff using VIM and getting the imputed values into the

bmi_imp variable t.test(bmi, bmi_imp)

On Sun, Jul 1, 2012 at 8:44 AM, mworni < reply@reply.github.com

wrote:

I just realized that this is under the wrong label - this question relates to NSQIP and not esophageal cancer.

It's hard to explain since it comes out of VIM. When you impute bmi using kNN, then you get for every patient a bmi value and an additional variable called bmi__imp which indicates for each individual if bmi was imputed or original.

I now wanted to compare bmi from the original dataset with the bmi values I got after the imputation. But maybe this is not possible.

On Sun, Jul 1, 2012 at 2:41 PM, Ricardo Pietrobon < reply@reply.github.com

wrote:

i'm still not entirely sure i understand the question, but:

  1. if you have two variables (bmi and bmi_imp) you should run a t.test using something like t.test(bmi, bmi_imp) , but i think you've done this before. is your question whether the test should be paired or unpaired? sorry, i still don't think i understand it
  2. if you are trying to compare the two datasets as a whole (all imputed variables compared at once) then you would have to compare beta coefficients and CIs, but i think you also know that

On Sun, Jul 1, 2012 at 7:48 AM, mworni < reply@reply.github.com

wrote:

the question is much more technical...

if you check the script - there is a ton of stuff new in relation to the imputation. It definitely works with the sample set - and I can make also a ton of comparisons between the original and the old dataset using VIM. Very nice graphs and all as code now in the script.

However, what if I would like to make a t-test between the imputed and non imputed bmi? How do I have to code this.

Can I run a t-test between the nsqip.sample and the nsqip.sample_IMPUTED dataset? Do I have to make a t-test between bmi original and bmi imputed using bmi__imp as an indicator variable? However, this would give me two different results!

Awesome stuff!!!!! Looks great!

On Sun, Jul 1, 2012 at 1:42 PM, Ricardo Pietrobon < reply@reply.github.com

wrote:

i'm not sure i understand the question, but:

  1. graphically through the VIM functions to check where the location is
  2. through summary measures and inferential tests (t-tests) to see whether the two are different
  3. comparing models adjusted for one vs. the other

you already know all of this, which tells me that you are asking something different

On Sun, Jul 1, 2012 at 5:21 AM, mworni < reply@reply.github.com

wrote:

I created now a nsqip.sample and a nsqip.sample_IMPUTED dataset. How can I compare bmi in the first with the latter dataset?

variable name is bmi in both datasets. bmi__imp is a variable that indicates if the variable was imputed or not (coded 0 and 1).

Reply to this email directly or view it on GitHub: https://github.com/rpietro/EsophagealCancerT1N0/issues/6

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6692811

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6692852

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6693149

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6693177

mworni commented 12 years ago

This won't work - I think I have to show you this, it will get clear what the problem is in a second.

On Sun, Jul 1, 2012 at 2:51 PM, Ricardo Pietrobon < reply@reply.github.com

wrote:

got it now. here is some mock code:

bmi_imp <- bmi

do your imputation stuff using VIM and getting the imputed values into the

bmi_imp variable t.test(bmi, bmi_imp)

On Sun, Jul 1, 2012 at 8:44 AM, mworni < reply@reply.github.com

wrote:

I just realized that this is under the wrong label - this question relates to NSQIP and not esophageal cancer.

It's hard to explain since it comes out of VIM. When you impute bmi using kNN, then you get for every patient a bmi value and an additional variable called bmi__imp which indicates for each individual if bmi was imputed or original.

I now wanted to compare bmi from the original dataset with the bmi values I got after the imputation. But maybe this is not possible.

On Sun, Jul 1, 2012 at 2:41 PM, Ricardo Pietrobon < reply@reply.github.com

wrote:

i'm still not entirely sure i understand the question, but:

  1. if you have two variables (bmi and bmi_imp) you should run a t.test using something like t.test(bmi, bmi_imp) , but i think you've done this before. is your question whether the test should be paired or unpaired? sorry, i still don't think i understand it
  2. if you are trying to compare the two datasets as a whole (all imputed variables compared at once) then you would have to compare beta coefficients and CIs, but i think you also know that

On Sun, Jul 1, 2012 at 7:48 AM, mworni < reply@reply.github.com

wrote:

the question is much more technical...

if you check the script - there is a ton of stuff new in relation to the imputation. It definitely works with the sample set - and I can make also a ton of comparisons between the original and the old dataset using VIM. Very nice graphs and all as code now in the script.

However, what if I would like to make a t-test between the imputed and non imputed bmi? How do I have to code this.

Can I run a t-test between the nsqip.sample and the nsqip.sample_IMPUTED dataset? Do I have to make a t-test between bmi original and bmi imputed using bmi__imp as an indicator variable? However, this would give me two different results!

Awesome stuff!!!!! Looks great!

On Sun, Jul 1, 2012 at 1:42 PM, Ricardo Pietrobon < reply@reply.github.com

wrote:

i'm not sure i understand the question, but:

  1. graphically through the VIM functions to check where the location is
  2. through summary measures and inferential tests (t-tests) to see whether the two are different
  3. comparing models adjusted for one vs. the other

you already know all of this, which tells me that you are asking something different

On Sun, Jul 1, 2012 at 5:21 AM, mworni < reply@reply.github.com

wrote:

I created now a nsqip.sample and a nsqip.sample_IMPUTED dataset. How can I compare bmi in the first with the latter dataset?

variable name is bmi in both datasets. bmi__imp is a variable that indicates if the variable was imputed or not (coded 0 and 1).

Reply to this email directly or view it on GitHub: https://github.com/rpietro/EsophagealCancerT1N0/issues/6

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6692811

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6692852

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6693149

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6693177

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6693227

rpietro commented 12 years ago

just to document our hangout conversation:

bmi_old <- bmi

do the imputation stuff into bmi

t.test(bmi_old, bmi)

On Sun, Jul 1, 2012 at 8:53 AM, mworni < reply@reply.github.com

wrote:

This won't work - I think I have to show you this, it will get clear what the problem is in a second.

On Sun, Jul 1, 2012 at 2:51 PM, Ricardo Pietrobon < reply@reply.github.com

wrote:

got it now. here is some mock code:

bmi_imp <- bmi

do your imputation stuff using VIM and getting the imputed values into

the bmi_imp variable t.test(bmi, bmi_imp)

On Sun, Jul 1, 2012 at 8:44 AM, mworni < reply@reply.github.com

wrote:

I just realized that this is under the wrong label - this question relates to NSQIP and not esophageal cancer.

It's hard to explain since it comes out of VIM. When you impute bmi using kNN, then you get for every patient a bmi value and an additional variable called bmi__imp which indicates for each individual if bmi was imputed or original.

I now wanted to compare bmi from the original dataset with the bmi values I got after the imputation. But maybe this is not possible.

On Sun, Jul 1, 2012 at 2:41 PM, Ricardo Pietrobon < reply@reply.github.com

wrote:

i'm still not entirely sure i understand the question, but:

  1. if you have two variables (bmi and bmi_imp) you should run a t.test using something like t.test(bmi, bmi_imp) , but i think you've done this before. is your question whether the test should be paired or unpaired? sorry, i still don't think i understand it
  2. if you are trying to compare the two datasets as a whole (all imputed variables compared at once) then you would have to compare beta coefficients and CIs, but i think you also know that

On Sun, Jul 1, 2012 at 7:48 AM, mworni < reply@reply.github.com

wrote:

the question is much more technical...

if you check the script - there is a ton of stuff new in relation to the imputation. It definitely works with the sample set - and I can make also a ton of comparisons between the original and the old dataset using VIM. Very nice graphs and all as code now in the script.

However, what if I would like to make a t-test between the imputed and non imputed bmi? How do I have to code this.

Can I run a t-test between the nsqip.sample and the nsqip.sample_IMPUTED dataset? Do I have to make a t-test between bmi original and bmi imputed using bmi__imp as an indicator variable? However, this would give me two different results!

Awesome stuff!!!!! Looks great!

On Sun, Jul 1, 2012 at 1:42 PM, Ricardo Pietrobon < reply@reply.github.com

wrote:

i'm not sure i understand the question, but:

  1. graphically through the VIM functions to check where the location is
  2. through summary measures and inferential tests (t-tests) to see whether the two are different
  3. comparing models adjusted for one vs. the other

you already know all of this, which tells me that you are asking something different

On Sun, Jul 1, 2012 at 5:21 AM, mworni < reply@reply.github.com

wrote:

I created now a nsqip.sample and a nsqip.sample_IMPUTED dataset. How can I compare bmi in the first with the latter dataset?

variable name is bmi in both datasets. bmi__imp is a variable that indicates if the variable was imputed or not (coded 0 and 1).

Reply to this email directly or view it on GitHub: https://github.com/rpietro/EsophagealCancerT1N0/issues/6

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6692811

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6692852

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6693149

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6693177

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6693227

Reply to this email directly or view it on GitHub:

https://github.com/rpietro/EsophagealCancerT1N0/issues/6#issuecomment-6693235