Serial CorelationI have a data set in which multiple values were collected from same patient. There are 25 patients but each patient would have about 30 observations.
These patients have their transcutaneous CO2 is measured at different time points.. And we are comparing different characteristics (oxygen requirement, Respiratory support) with normal and abnormal transcutaneous CO2 measurements. Also correlating transcutaneous CO2 with FiO2. We will also be doing multi variate regression analysis.
This obviously leads to problem of non- independent observations and error in the P value by standard tests.
I was talking to a statistician who analyses the data using SAS . He mentioned that he can adjust the statistical test for serial correlation in SAS.
is it possible in jmp to do that ?
Re: Serial Corelation xtreg command in STATA program
He used regression analysis for panel data.
According to statistician, "simple t test is not applicable for the data".. I would still want to replicate the results in JMP
Re: Serial CorelationHmm I am unable to reach him. But will ask
Meanwhile I am trying to figure out if Fit Y by X is a appropriate strategy in this case ?
Re: Serial Corelation"Statistician just send me the results of his analysis by SAS". Can you ask them to also send the SAS code they used and post it here (not the data), please?
Re: Serial CorelationStatistician just send me the results of his analysis by SAS.His analysis of the mean values of FiO2 for normal and abnormal transcutaneous CO2 is our much different than what I calculated based on simple univariate fit Y by X comparison on jmp.
The only explanation I can give is it has something to do with serial correlation of the values (multiple values recorded every 30 minutes for each patient). In jmp I have counted each row as a independent observation. I am not entirely sure what the statistician has done (I am unable to reach him).
I don't want to take his analysis on face value and want to independently verify the results (also helps me understand it when I'm writing)
Re: Serial CorelationSo you have about 30 observations for each of 25 patients. Each observation is a time point that includes patient ID, transcutaneous CO2, FiO2, oxygen requirement, respiratory support. Is this much correct?
Yes this is correct
I understand by 'serial correlation' that you mean the response is auto-correlated. Do you know why? For example, are the measurements taken over a period of time that is too short for the random variation, or is there a treatment over the same time period that affects the response?
Measurements are taken every 30 minutes. These patients are in the hospital getting various sort of treatment so various treatments over the time period can affect the response.
Which of these variables is the response or outcome that is important to you?
Transcutaneous CO2 is the predictor variable (classified as normal and abnormal based on the values) and predicted variables are FiO2, flow rate, bronchiolitis score and pew sco most important are the FiO2 and the flow rate re.
Are any other variables, such as age, gender, or disease state?
gender and patient's weight are other variables that can impact
Is the normal or abnormal transcutaneous CO2 determined independently for each observation?
Yes that is determined independently for each observation
Did the statistician explain the nature of the adjustment that would be used?
Not really
I think that a mixed model with patient as a random effect would be OK, but it might be necessary to nest the other parameters for a random coefficient model. JMP can perform some of these regressions. JMP Pro might be required for the other regressions.
Re: Serial CorelationSo you have about 30 observations for each of 25 patients. Each observation is a time point that includes patient ID, transcutaneous CO2, FiO2, oxygen requirement, respiratory support. Is this much correct?
I understand by 'serial correlation' that you mean the response is auto-correlated. Do you know why? For example, are the measurements taken over a period of time that is too short for the random variation, or is there a treatment over the same time period that affects the response?
Which of these variables is the response or outcome that is important to you?
Are there any other variables, such as age, gender, or disease state?
Is the normal or abnormal transcutaneous CO2 determined independently for each observation?
Did the statistician explain the nature of the adjustment that would be used?
I think that a mixed model with patient as a random effect would be OK, but it might be necessary to nest the other parameters for a random coefficient model. JMP can perform some of these regressions. JMP Pro might be required for the other regressions.
