range for c-hat

[tpinn]

Is there a rule-of-thumb for the acceptable range of c-hat values? I'm particularly interested in a lower limit, as c-hat values greater than 1 can be accounted for (to some degree at least) by using QAIC. But is there a lower limit for what is acceptable? Thanks!

Tracy

[Sunday]

Hi tpinn,
I was just reading the "Program MARK: A Gentle Introduction" because I have some questions about c-hat as well. I saw your post earlier and thought this might be useful to you:

"What if ˆ c is < 1?
What if ˆ c < 1? In the preceding, we mention the case where ˆ c > 1, indicating some degree of lack of
fit, reecting (in all likelihood) overdispersion in the data. Now, if instead, ˆ c < 1, then we generally
consider this as reecting underdispersion. While the intuitive thing to do is to simply enter the ˆ c
as estimated (discussed below), there is lack of unanimity on how to handle ˆ c < 1. Some authors
recommend using the estimated ˆ c, regardless of whether or not it is > 1 or < 1. However, still others
suggest that if ˆc < 1, then you should set ˆc = 1 (i.e., make no adjustment to various metrics). For the
moment, the jury is out - all we can recommend at this stage is - if ˆ c > 1, then adjust. If ˆ c < 1, then set
ˆc = 1, and ‘hold your nose’."

Trevor

[tpinn]

Thanks Trevor,
So perhaps there is no real rule-of-thumb for a range of c-hat? But it sounds like if c-hat is close to 0, that is probably too low (based on some other posts on this forum). So I am wondering about c-hat = 0.2 or 0.3? But maybe this depends on other factors! Thanks again,

Tracy