Teratology Study: Test for Trend in Correlated Data

The data in the following table are reprinted from Bradstreet and Liss (1995). There are 96 total litters (clusters), with a total of 1248 fetuses. Summing over the litters, we observe an average malformation rate of 4/288 (1.4%) in the control group, 6/327 (1.8%) in the 8 mg/kg group, 8/305 (2.6%) in the 80 mg/kg group, and 10/328 (3.0%) in the 800 mg/kg dose group. Does the increasing average empirical malformation rate provide evidence of a

dose effect? To obtain an accurate answer, we must first account for the correlation within litters, as we cannot generally assume that offsprings from the same mother are independent. Moreover, while the number of clusters and the total sample size are relatively large, the number of affected fetuses is relatively small.



Analysis of the above data in StatXact produce the following results:

Conclusion: notice that the exact p-value is about half of asymptotic p-value. The p-value indicates that the trend in response for increasing dose levels is not significant at 5% level of significance.