The effects of serial dependency on the use of Chi-square for analysing data in dyadic interactions

Author(s): Hartmann, D., Mitchell, C.

Abstract: Chi-square tests have been recommended as a method for assessing dependence between the sequential behaviors of interacting members of a dyad. The presence of serial dependence within the sequential behaviours of a single subject presumably violates an assumption of chi-square. In order to assess the effects of serial dependence an the Chi-square Test of Independence, we conducted 24 null hypothesis simulations, each involving behavior chains for 5.000 randomly sampled dyads. We manipulated level of within subject serial dependence, base rates of the target behaviour, and the length of behaviour chains. The results indicated that the probability of Type I errors (a) increased dramatically with increasing levels of the absolute value of serial dependence, (b) varied ac a function of the interaction between base rates and serial dependence, and (c) remained unaffected by the length of behaviour chains. The Chi-square lest of Independence should not be used inferentially when levels of serial dependence exceed the absolute value of 50. The test should he used inferentially only with caution when moderate levels of serial dependence are present.

Zurück | Weiter