1. In categories (outside courses and groups) "User" means "User who participated the test", in test they are called "Participants".
2. In courses and groups "User" means "Member of Course" or "Member of Group".

Screenshot of Problem

There are 5 members in course where test is located, 3 of them finished the test:

• a.meierhans and b.muster are course members, which didn't start the test.
• b.sommer, l.blum3 and o.witzig finished the test (2%, 97%, 16%).

ø Percentage / User summarize all percentage of participants which started/finished the test (2% + 97% + 16%) and divide it with numbers of course members (115% : 5 = 23%).

• 2 persons didn't start the test and this value assume that they haven't passed the test (0% correct answers).
• outside a course/group, value uses only "real" participants of test.

VARIANT 1

To avoid misunderstandings between "User", "Participants" and "Members" we propose following changes:

1. Column "ø Time Spent / User" and "ø Percentage / User" should be renamed to "ø Time Spent / Participant" and "ø Percentage / Participant".
2. Both columns count only "real" participant values (persons who didn't started test, aren't included).
3. An additional column "ø Percentage / Members" will be introduced if test is located in courses and groups. This value calculates percentage of ALL members (like now).

• An additional column "ø Percentage / Participants" will only count "real" participant values (persons who didn't start the test, aren't included).

Heyser, Björn [bheyser], 30. March 2017:

One could say - users / participants / members and than we still have terms like student or examinee in the heads of our end users - we should streamline this and choose a single term as label used in every context we are talking about them. The answer should be NO, because the existance of precise terms offers the opportunity to just precisely separate a subset of people from another set of people in mind.

Everybody knows what we are talking about in the context of ilias, when we label anything with users. So this term should not be abendoned as it is our basic term, that can be used as the overall set of people for every context dynamically. we dont need to separate local users from global ones, when we just mean the parent set of people for a members set, they are simply users - means, all users - from current context. Users are managed behind a wall from the viewpoint of any end user in the repository, so they are there, we dont know where they come from. the first group of people i can define anywhere is always a more precise subset than the context sensitive overall set of people called users.

The issue shows itself just when we deal with these different sets/subsets of people in one screen for the same thing. And exactly here the advantages of the precise meaning behind such a people's subset term perform at its best. Example: Do you want the avarage result to be calculate based on "actual participants" or "all <objtype> members"? This could of course also help a radio group filter to be more compact and handy in the manner of understanding.

What I want to say is, that I strogly second the arguments in this request as we have these terms in use but do not use them the right way consistently. This leads to more misunderstandings that we would have open questions, when we just used users as term for all places (!)

I do not fully second the suggestions to solve the problem. Although they fully clear up the things for the end users, it might be useless information that is shown there.

Perhaps we can let the test creator/author decide wether this test should consider the actual participants as dynamically valid "user set" or the course/group/anyParent members by adding an additional setting for this. The statistic can automatically use the correct overall set of people and calculate/show the information accordingly. Labels should be choosed dynanmically of course. Either the test provide the language variable him self, or he asks the corresponding parent object type for a language variable to be used.

For the situation of using parent members as overall user set for the test a filter can hide the members that did not start yet for a general statistic about performance, when they are shown, it is more a list of obligatory participations that shows the test scenario's progress. For the aggregated statistic we could also offer this filter, or we should just left out the members that did not start yet as this should be a statistical measurement for quality only i guess. When we consider users that did not start yet for this calculation, it is not clearly identifyable wether the fact of not yet happened starts or a bad performance leads to a small average.

Additional Suggestion: When we consider to improve the use of the precise terms for people in general, we should consider to differ between invitees and participants. The fixed participants mode for the test offers the same advantages that we get with a members container above, that means we also have to deal with the same conceptional issue here, too.

AT 2017-04-03: I would prefer Varaint 2. Ther term "user" should be replaced by "participant". The third coulumn proposed in Variant 1 will produce highly skewed numers without adding interpretational value. 
Averages within groups or courses must not use the number of members but the number of participants as denominator.