Time-dependent transition probabilities and the assessment of seasonal effects on within-day variations in chewing behaviour of housed sheep

State transitions in the chewing behaviour of six half-breed (Ile de France X Texel) yearling female sheep (Ovis aries L.) were studied by using jaw m ovements recorded continuously over 5 days at the end of a number of experi mental periods from 21 September 1992 to 4 April 1993. The sheep were house d in individual pens. Each of them received the same diet, that is, 250 g/d ay of concentrate mix (15.5% crude protein (CP), 36.5% neutral detergent fi bre (NDF)) fed at 0900 h and natural grass hay (6.7% CP, 69.1% NDF) fed ad libitum at 0915 and 1600 h. Mineral salt blocks and water were continuously available. The main objective was to assess seasonal effects on within-day variations in the chewing behaviour of sheep, at small to large time scale s within a day. We therefore focused on two experimental periods characteri sed by contrasting conditions of daylength and temperature(i.e., 'Period 1' : 610 min daylight, mean temperature of 10.9 degrees C, and 'Period 4': 550 min daylight, mean temperature of 7.2 degrees C). In particular, differenc es between periods in the nycterohemeral pattern of chewing behaviour and t he quality of forecasts of chewing states were tested. We submitted our dat a to a new method of analysis that we developed: the method of time-depende nt transition probabilities, and compared the results to those obtained usi ng other methods that were available in the literature. Overall, the sheep spent more time eating in Period 1 than in Period 4. Spe cifically, a secondary peak in eating activity, which was observed in the e arly afternoon in Period 1, was absent in Period 4. The nycterohemeral patt ern of eating activity showed significant differences between periods, at t he main rhythmic component of 24 h and at short components around 2 h. Such differences were not observed for ruminating and idling activities. The qu ality of forecasts of chewing states decreased from Periods 1 to 4, in term s of accuracy (based on R-2) and lead of reliable forecasts (i.e., \forecas t - observation\ < 0.1). The most (least) accurate forecasts were obtained for the ruminating (eating) state in both periods. We have attributed the d ifferences that we found between periods to daylength instead of temperatur e because the sheep were mostly within the thermoneutral zone in our study. By comparison, using mean hourly times of eating activity, significant dif ferences between periods were detected for the 24-h rhythmic component and the 4-h component, instead of the 2-h component, probably because of aliasi ng (i.e., when the sampling time interval used is longer than suited; the m inute was found to be a suitable interval length in the calculation of time -dependent transition probabilities). Using the age-dependent model of Rook and Penning, minor differences between periods were detected. On that basi s, the method of time-dependent transition probabilities may be brought for ward as a complement of value to existing methods of behavioural data analy sis. (C) 2000 Elsevier Science B.V. All rights reserved.