The nature of the situated action of the INFO event is studied via the relationship between the occurrences of the INFO event and several contextual, explanatory variables in the collaborative hypertext database. The occurrences of the INFO event are likely to indicate an increasing concern of some social and coordination issues of an ongoing collaborative process. The resultant model of the context-behaviour relationship is developed within the framework of the study so as to provide empirical evidence for the evaluation of existing models of collaborative work in CSCW studies.
Overall Usage Profile of Collaborative Hypertext Databases
The use of the five collaborative hypertext databases in the MUCH system resulted in thousands of transactions during the three-month longitudinal observation. The scale of use for each database varies. The class database was accessed by the largest group of users (User = 30) with the widest spectrum of functions from the presentation layer (Action = 25). Table 4.1 summarises the overall scale of use for each database, in terms of the total number of transactions observed, the total number of unique nodes involved, the size of the user pool, and the number of distinct functions required.
Table 4.1. Usage of the five collaborative hypertext databases over 3 consecutive months.
|Groupware||12089||750||11||22||Tue Apr 19||Mon July 25|
|Class||9247||474||30||25||Wen Apr 20||Mon July 15|
|Management||7068||698||16||22||Wen Apr 20||Thu July 21|
|OSCAR||3148||315||7||18||Tue Apr 19||Mon July 25|
|General||1590||96||10||21||Tue Apr 19||Fri July 15|
The percentage use of each function available in the presentation layer is given in the appendix for each database (Table A.1). Some heavily used functions appear to entail some cognitive and behavioural overheads due to particular design decisions for the presentation layer of the MUCH system. For example, the percentage use of the UNFOLD function is predominant in all cases but the General database. On the other hand, a heavily used UNFOLD function in a particular database may indicate that the database has been mainly used for navigating and reading tasks rather than substantive authoring tasks.
The average percentage usage of the INFO function is about 1% over the five databases. It is rather small in contrast to the percentage use of functions such as UNFOLD, READING, and LOCK. The dispersed occurrences of the INFO event suggest that the INFO function is only required for some needs to do with higher degrees of coordination coupling in collaborative writing. The independent working mode is primarily supported by the MUCH system as the system is designed to facilitate asynchronous collaborative writing. Although it is difficult to define the range of tasks to be enclosed by the independent mode, it appears that tasks can be fulfilled via the INFO function go beyond this range.
Awareness Behaviours and Contextual Characteristics
The occurrences of some rare event in the world are generated by Poisson processes. The probability distribution of the occurrences of the event associated with a Poisson process is called a Poisson distribution. The Kolmogorov-Smirnov (K-S) goodness-of-fit test of the null hypothesis, that the data are drawn from a population with a Poisson distribution, indicated that the occurrences of the INFO events follow a Poisson distribution with the mean = .0308 (K-S Z = 7694, p = 5946 two-tailed). Table 4.2 shows the occurrences of the INFO event observed in one-hour time intervals.
Table 4.2. The occurrences of the INFO event in one-hour time intervals in using the class database
In the vast majority of the time intervals observed, INFO events did not occur at all. In 19 time intervals, the INFO event occurred once. The higher the count of the INFO event in an interval, the fewer number of such intervals can be found, except one case (Occurrences = 6). The highest count of the INFO event for the class database is 7.
Several Poisson regression models were fitted to the time series data of the occurrences of the INFO event and corresponding contextual characteristics of the awareness-seeking path of users in a given time interval. In particular, the model specified as follows resulted in an adequate goodness-of-fit statistic for the data:
ln(count) =0 + 1*Zcredit +2*Zlevel +3*Zlink +4*Zword +5*Zsize +6*hour
The explanatory variables, or independent variables, were specified as the covariant variables in the model, except the hour variable, which is categorical. The GENLOG procedure in SPSS revealed that the model adequately represents the data set (Pearson 2= .0960, DF= 1, P = 76). The Poisson distribution of the count of the INFO event was specified in the model. Table 4.3 shows the results of the estimates for the model and associated significance tests.
Table 4.3. A Poisson regression model of the relationship between INFO events and contextual characters of the path activated in the same time intervals. (Source Database: Class; Observation: April 20 - July 15, 1994).
|xk||k|| ||*p < 0.05||Effect|
For a given explanatory variable x, a positive sign of the estimate suggests the likelihood of the INFO event count increases with the level of x, with the other xs held constant. A negative sign suggests that the likelihood of the INFO event count decreases with the level of x. An insignificant test at a conventional level such as 0.05 suggests that the effect of an x on the response variable is not statistically different from zero.
Three estimates in this model are statistically different from zero. Among these, Zlevel has a positive coefficient, whereas both Zcredit and Zsize have negative coefficients. For example, the estimate of level= 19.28 (p < 0.05) for Zlevel suggests that the occurrences of the INFO event in a given time interval are likely to increase as the Zlevel in the same interval increases.
The Poisson regression model can also be interpreted in terms of the marginal effect of an x on the expected INFO count. Liao (1994) discusses five ways of interpretation generalised linear models, including the two ways utilised in this paper. The marginal effect of xk on expected y is given by k, where = 0.0308 is the expected occurrences of the INFO event in a given time interval.
The marginal effect of Zlevel on the INFO count is (0.0308)(19.2786) = 0.5938. In other words, if the value of Zlevel increases by one, the expected INFO count will increase by 0.5938, other things being equal. In contrast, the marginal effect of Zcredit is (0.0308)(-11.2120) = -0.3453, suggesting that the expected INFO count will decrease by 0.3453 as the Zcredit increases by one standard deviation.
Preliminary analyses found the contextual variable Zlevel is negatively correlated with the frequency of UNFOLD and it is positively correlated with the frequency of LOCK (Pearson's r= -0.529, p < 0.001 and r = 0.187, p = 0.001, respectively).
Finally, if a dummy variable Phase is included in the model to specify the two phases of reading and authoring, the Phase variable would have the most predominant effect on expected occurrences of the INFO. The writing phase considerably increases the probability of a higher INFO count. The frequency of the INFO events was indeed increased when group essays were emerging in the MUCH system. Students were more interested in new nodes coming into the shared repository.