### Conclusions

#### Posted by zebra77 on 23 Jun 2008 at 14:52 GMT

The viability trade-off between hatching success and survival seems to operate only for the first two categories of egg order, i.e. the first versus the middle one. Only those two categories show statistically significant differences in both measures of viability and they differ in the direction of the difference.
How to deal with that?

Given the incomplete data set, dividing eggs into the three categories is very reasonable and justified. However, the approach of contrast analysis does not allow one to explore fully the predicted directionality of change between the first, middle and last category. Would it be possible to treat category as a continuous variable?

### RE: Conclusions

#### kwsockman replied to zebra77 on 04 Jul 2008 at 21:56 GMT

Hi Joanna,

Thank you very much for initiating this discussion on my paper. I am sorry for the delay in responding, but I am in the field and have only occasional internet access.

Yes, I agree that the trade-off is most apparent for the comparison between first and middle-laid offspring. As I mention in the paper, one reason the last-laid may seem to suffer low hatching success compared to middle laid might be due to a shift from incubation to feeding behavior once the first eggs hatch. But I do not know this with certainty. It is simply a hypothesis.

Regarding treating hatching (or laying) order as a continuous variable, in fact, I had done this in the first two submissions of the manuscript. But, after careful consideration, I changed it to a categorical variable. To make it continuous initially, I simply replaced first-, middle-, and last-hatched with the values 1, 2, and 3, respectively. However, this seemed arbitrary to me. Does being middle-hatched result in twice the effect and does being last-hatched result in three times the effect as being first-hatched? Moreover, to effectively model the inverse U-shape in the hatching order graph, I had to introduce a quadratic term. However, how does one square a value of "middle?" In the end, it seemed the conservative and sensible approach was to model it as categorical, despite the limitations you raise.