One factor that determines the efficiency of collaborative learning is the composition of the group. This factor is defined by several variables: the age and levels of participants, the size of the group, the difference between group members, etc.

Regarding the number of members, small groups seems to function better than large groups in which some members tend be 'asleep' or excluded from interesting interactions. Most of the mechanisms described in the previous section, e.g. mutual regulation, social grounding, shared cognitive load, ..., can only occur between a few participants. This does not argue in disfavor of large group sessions. It simply means that distance learning activities should also include 'closed' sessions, in which a restricted number of subjects collaborate and/or 'monitored' session in which the teacher takes care that no learner is left out the interaction.

Regarding the participants, some developmental level is necessary to be able to collaborate, but this is only an issue for children and does hence not directly concern current distance education activities which mainly concern adult learners.

The most intensively studied variable is the heterogeneity of the group. It refers to the objective or the subjective differences (how subjects perceive each other) among group members. These differences can be general (age, intelligence, development, school performance, ...) or task specific. Results indicate there exists some 'optimal heterogeneity', i.e. some difference of viewpoints is required to trigger interactions, but within the boundaries of mutual interest and intelligibility. Heterogeneity can easily be understood as a condition to trigger conflicts and require social grounding, two important mechanisms described above. Heterogeneity is also implicit in the socio-cultural theory and its related mechanisms (internalization and appropriation) which rely on the observation of adult-child pairs or at least pairs with one member being more knowledgeable on the task than the other.

Internet-based information and communication tools have a great potential with respect to heterogeneity: no infrastructure can better cross geographic, cultural and professional boundaries. Nevertheless, human beings have a natural trend to assemble with those who are the most similar to them. When participants join the group on their own decision, there is no control of heterogeneity. If the tutor observes too much homogeneity among the group members, he may modify some conditions in order to activate anyway the mechanisms that normally rely on heterogeneity. He may for instance allocate role to participants which will inevitably create conflict or provide them with contradictory information.

The effects of collaboration vary according to the task. Some tasks prevent the activation of the mechanisms described above, while other tasks are appropriated. For instance, some tasks are inherently distributed and lead group members to work on their own, independently from each other. Interaction occurs when assembling partial results, but not during each individual's reasoning process. Without interaction, none of the described mechanisms can be activated. Some tasks are so straightforward that they do not leave any opportunity for disagreement or misunderstanding. Some tasks do not involve any planning and hence create no need for mutual regulation. Some tasks cannot be shared, because they rely on processes (e.g. perception) which are not open to introspection or on skills (e.g. motor skills) that leave no time for interaction.

If distance teachers want to take these features into account, a first attitude would be to use only collaborative learning for tasks for which it will get its optimal efficiency. Another solution is to modify the task, as explained in the previous paragraph, to make them more suited for collaboration. For instance, the 'jigsaw' method consists of providing group members with partial data. This method artificially turns a monolithic problem into a task which requires collaboration.

Task features also include the environment in which the task has to be performed. This is especially important in computer-based tasks. The software features may modify interactions among learners. For instance, if a computer-based task provides the learner immediately with a feed-back on their actions, it may prevent them to discuss the consequences of their action

Whatever task and group members have been selected, the collaboration may not work because the medium used for communication is not adequate. It would be beyond the scope of this paper to describe each available media. Basically, most of current widely available Internet-based tools use text-based communication, synchronous or asynchronous, with mostly fixed graphics and images. Voice and video interaction or voice and video mail are of course available, but the overload of standard networks and the limits of currently available hardware has postponed their larger use in current distance education.

Most of the mechanisms described in the previous section can be conveyed via text-based communication, but with some perturbations. For instance, the cost of interaction being higher with text, the group members may reduce the number of disambiguating sub-dialogues used in social-grounding. At the opposite, in asynchronous text messages, they have more time to build sentences which are less ambiguous. Without video link, members also loose facial expressions which are useful to monitor the partner's understanding. Even with video images, they may see their partner but ignore where the partner looks, something which is important for understanding what she refers to. Some video system support eye contact with appear to be related to metacognitive aspects.