Added Individual Assignment
You can submit an individual Word document (no more than four A4 pages) in which you shortly describe a more thorough and sophisticated statistical analysis of the data you collected for your individual robot.
The document should be titled “Added Individual Assignment” and contain at least the sequence of behavioural activities you collected from your “focal” robot, both as raw data and summarised in a transition matrix.
The methods for behaviour analysis are covered in the slides and accompanying Excel Workbook “Analysis of Robot Behaviour” and the spreadsheet “seqdep” (plus guidelines) (Course Work Assignments > Robotics > CW3 > Supporting Documentation & Tools).
The workbook “Analysis of Robot Behaviour” consists of four spreadsheets that contain valuable information about the chi square test for dependency between behaviour acts and similarity (distance) between behaviour profiles as used in behaviour analysis and how to compute them.
The first sheet (“Example Light Signal vs Movement”) illustrates the first 25 slides of the accompanying set of PowerPoint slides “Analysis of Robot Behaviour” and deals with the setup of a simple chi square test for associations between (behavioural) events. By clicking at the coloured cells you can see the functions that have been used for the computations.
The second sheet (“Example Chi Sq Distribution”) contains tables and plots of the Chi Square Distribution and exemplifies the issues of statistical significance as mentioned on slides 15 – 25.
The sheet “Example Sequential Analysis” is not only of use for this additional assignment but should also be helpful for the analysis of robot data you are required to carry out for the main part of CW3. By clicking on the green cells you can uncover the choice and use of functions to generate transition matrices which you can then adapt and apply to the data collected by the group. The sheet also carries out a simple chi square test for sequential dependency, but this test does NOT bypass diagonal zero’s; for a procedure that does take this in account you have to use the spreadsheet “seqdep”. A comparison between the results of the “naïve” and sophisticated (seqdep) test (applied to data of rat behaviour) is the topic of the fourth spreadsheet (“Chi Sq Rat Data”) and refers to slides 57 - 63.
Finally, sheet 5 (“Similarity Rat Data Profiles”) illustrated to compute Euclidean and Correlation Distances as they are used in Cluster Analysis. Once computed, you could a cluster analysis by hand (if you do not how to do this automatically in R or MatLab) given the fact that the number of behavioural elements you are dealing should be quite small.
A note on the use of “seqdep”.
“seqdep” (plus guidelines) is a spreadsheet for doing a Chi Square test bypassing diagonal zero’s. Read the “guidelines” for a description and explanation of its use. Be aware that the spreadsheet was set up in an older version of Excel and depends on the use of macros, therefore do not convert it in a newer version of Excel.
Suggestions for things to include in the assignment
Choose one or more of the topics below
I. A Sequence Plot of the activities (0.5 mark).
II. Combining activities with the similar profiles (as assessed from a correlation matrix, a PCA or a cluster analysis) (3 marks).
III. Chi Square Tests (bypassing diagonal zero’s) to check whether the frequencies of transitions are statistically independent (3 marks).
IV. If they are not:
· An identification of those elements that follow each other sequentially
more than expected by chance (0.5 mark)
· Possible solutions to deal with overt dependency (0.5 mark)
· A Markov State Space diagram based on the transition rates after
correcting for dependency (0.5 mark)
V. A Markov State Space diagram for various episodes (e.g. begin- and end of the experiment (2 marks).