8

I have serial hematological measurements data and I have plotted their means and SE in Stata. On the y-axis I have for example hemoglobin and time (visit days) on the x-axis hence I can visualize hemoglobin levels with time (whether it is decreasing or in increasing). The level decreases up to sometime and increases again. What test can use to test whether this is significant or not? Stata and R related answers are welcomed.

Thank you.

Julie

mpiktas
  • 33,140
  • 5
  • 82
  • 138
Julie
  • 533
  • 1
  • 5
  • 7

1 Answers1

3

It seems that your problem can be stated as change-point problem. R packages dealing with such type of problems are segmented and strucchange. Since you want to look into changes in time trend (and time trends always need special treatment in linear regression), I suggest differencing your hemoglobin level data and then testing whether there is a change in mean.

Look also into answers for this question.

Community
  • 1
mpiktas
  • 33,140
  • 5
  • 82
  • 138
  • Thank you for the response. I had initially used linear regression to examine this, with which I tested the change in mean between tow time points. However, the challenge comes when testing for the general trend over all the time periods. That means having one significant (or non-significant) value. I believe there are several assumptions that have to be incorporated. I have also seen that non-parametric tests (Mann-Kendall) can be used to test for trend. I am not quite sure how feasible this will be with the data I have maybe for non-normal data.Will look at the packages suggested. Thanks. – Julie Jan 30 '11 at 09:57