43 
LORATORY SOFTWARE 
CONCLUSION: DEMANDS FOR FUTU 
5. Conclusion: Demands for Future Exploratory Software 
The explorations of job spells of the German Life History Study were confronted with 
the problem: how to see many variables in the course of time and to compare groups in 
an overall view? Additionally, I had to cope with the problem of censored data and the 
data mass itself. 
The presented graphics had been developed in two aspects. In the first part for 
a descriptive point of view: to inspect individuals and birth cohorts, to visualize a time 
axis, to describe censored and uncensored job durations, to contrast groups, and to 
describe persons vs. spells. I used matrix displays to raise dimensionality, combinations 
of plots, pattern matrices to visualize durations, and colour attributes to improve 
visibility. In the second part, the statistical survivor functions and hazard rate functions 
were considered and compared by groups. 
All of the plots had been constructed "manually", by a bunch of different 
computer programmes. Whereas most software packages have implemented modules 
for survival models, there is a lack of exploratory tools for this kind of data. First steps 
are done by STATA with the commands for survival data, and by Atkinson with an 
XLISP-STAT programme. With my paper presentations of static plots, I can only 
contribute some concepts to the exploratory work. More effort on visual tools by 
sostware developers are desirable. 
One of the most important tools, interactivity, is indicated by the design of a 
survival data exploration window (figures 11 and 12). The advantages of an interactive 
user interface are to change groups and variables very quickly, to combine different 
types of plots, and to see more dimensions by adding movement. A user-driven 
window design allows individual perception preferences. An additional option should 
integrate resampling methods. I used this mainly to build exploratory subsets of data, 
but it is also a technique for finding robust trends. 
The computer development of the last 20 years teaches, that "large" data sets of today 
will get smaller and "slow" programmes soon will become fast. Therefore graphic 
software achieves better and better performance. Thus, data exploration will be 
possible even with complex data structures. Visualization assists human mind to step 
from concrete data to abstract ideas, to go back to the concrete, and to gain new 
insight.