4

I have the following situation. We want to measure if animal weight has an influence on their home range size (measured as an area).

However, there is inherent error measuring an animal home range. The method we're using output the home range value plus a 95% confidence interval.

Is there a way to incorporate this confidence interval in the regression of home range and mass?

I am aware that you can add weight to samples but I'm not sure if there is a better way.

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
JMenezes
  • 427
  • 2
  • 7
  • 1
    Have you looked at Errors-In-Variables models? If I am understanding you correctly, confidence intervals wouldn't do what you are intending them to do. Maybe you should explain the calculation method that you are using to find the home range in the first place. – Dave Harris Jul 29 '20 at 20:42
  • 2
    @Dave An EIV model deals with errors in the *explanatory* variables. This problem appears to be of the form $Y_i=x_i\beta+\epsilon_i$ where $$\operatorname{Var}(\epsilon)=\sigma^2\mathbb{I}_n+\operatorname{Diag}(\sigma_1^2,\sigma_2^2,\ldots,\sigma_n^2)$$ and the *response* measurement errors $\sigma_i^2$ are specified (they do not need to be estimated). – whuber Jul 29 '20 at 20:51
  • @whuber, ah, I misunderstood the question. – Dave Harris Jul 29 '20 at 20:53
  • @whuber is correct, there is an error associated in measuring the $Y_i$, which has been estimated. I found a similar [question](https://stats.stackexchange.com/questions/235693/linear-model-where-the-data-has-uncertainty-using-r), and they seem to be suggesting adding weights to the explanatory variable proportional to the inverse of the variance $w_i = 1/\sigma_i^2$ – JMenezes Jul 30 '20 at 22:07
  • Unfortunately that doesn't work. It's wrong because the weights don't accommodate the contribution of the unknown value $\sigma^2,$ which often is substantial. – whuber Jul 31 '20 at 15:52

0 Answers0