4. Discussion

The experimental results showed that the ANOVA for task time and ranking were significant, unlike the ANOVA for mean number of errors which was insignificant.

The insignificance of mean number of errors is in contradiction with our hypothesis which stated that mean number errors was expected to be proportional to the design of the web page. It is possible that this contradiction can be attributed to:

  1. Limitations of our model: The pages tested were a simplification of existing web pages where links are usually impeded in text instead of being more of a menu style arrangement.
  2. The fact that the questions (tasks) were selected to be of low difficulty resulted in fewer errors by our subjects, thus, making the analysis of this variable difficult.

Thus we will emphasize our discussion on the ranking and task times which were significant.

From Table 1 and 2 it can be seen that a trend exists:

  1. Model 1 is both the most difficult to use and with the highest mean task time. This is due to the big depth (2x6) that model 1 has. Under investigating model 1 subjects had to browse 6 steps to reach the information, this is both time consuming and not favored by users.
  2. On the other hand model 4 seems to have the lowest total task time and is ranked second in preference. Model 4 is a 4x1-16x1 model (first page breadth = 4, second page breadth = 16).
  3. Even though model 3 has the smallest mean time it is ranked 4th in preference among the different models.
  4. Increase in breadth (in the range investigated) does not seem to effect the speed.

From points 1 and 2 above our original hypothesis that access time is proportional to the depth of the tree structure of the arrangement of the links of the web page is justified.

Point 4 above is proven by observing the fact that model 5 (16 links at level 1) has the 2nd biggest task time and model 4 (16 links at level 2) has the second smallest task time. This result proves that the number of links per page is not a factor when it comes to access speed. People might take more time to scan through the page but this is counter-balanced by the reduction of depth associated with the increase of links per page.

There are some anomalies while comparing Table 1 with Table 2.

Both of these anomalies are difficult to be explained scientifically. Both Model 3 and Model 5 are of two step depth but they seem to follow extreme opposite trends on the statistics. This might be due to the effect of breadth on the design. Model 5 (16 links on front page) as expected is slow, and model 3 with both of its levels in the "seven plus or minus two" category gives the smallest task times. On the other hand what seems to be an "anomaly" is the preference ranking given to these designs, which proves once again the diverse nature of people.

Our subjects, made positive comments about their experience in testing this model. Comments ranged from "I learned more about Cyprus" to very precise suggestions (by one of our subjects) on how this test should be run and what should be improved.

With the exception of only one (number 12) subjects followed carefully our instructions and carefully answered the 10 questions assigned to them. Subject 12 eventhough he attempted all questions he gave up after browsing (unsuccesfully) long enough for some questions. Maybe an addition to the introduction emphasizing that subjects should attempt and record answers for all questions would have prevented this incomplete record.

Overall our results are in agreement with those of Kiger (1984) where it has been proven that access time is proportional to depth in menu selection. Thus our assumption that the arrangement of links on a web page behaves like options in a menu is justified.


Continue
Return to the Title Page for "Depth vs Breadth in the Arrangement of Web Links"