The trouble with statistics is that even when you think you have the right one the data doesn't want to play ball! Correlation was productive but has not revealed anything too startling. There are some average correlations but nothing particularly strong that could not be anticipated. So the last couple of weeks have been investigating other techniques.
The one I am now looking at using is cluster analysis. This is because what I really want to know is how different variables gravitate towards each other. With this I can look at the individual students to get a better and more grounded picture. The problem has been that the data has to be in the right format to use cluster analysis in SPSS so I have have to amend the data to get it to work. Unfortunately the online help from SPSS is only partially helpful so thank goodness for the internet where others who have passed this way have posted their procedures.
It is all very frustrating at times as I don't really want to get deeply involved with advanced stats methods but that is the price I have to pay for not designing a simple investigation using basic techniques! Hopefully in a week or so I will have some facts to work with. I'd better as I have to present initial findings at a workshop in two weeks time with some of our tutors!
No rest for the wicked.
Monday 25 August 2008
Thursday 7 August 2008
The Right Statistic
Another two weeks on and progress has been made but frustratingly so. The latest questionnaire went out to the 41 tracking sample of students on 21st July with the request for them to be completed by 1 August. The date came and only 24 had responded, however, an email reminder to those who had not completed has now yielded 37 responses with a further one likely when the student returns from holiday.
That is the progress so what is the frustration. Needless to say that is the statistics. My intention of using ANOVA was scuppered once I started to look more closely at the detail. Originally I had a mix of free and categorised data. When I looked again at the frequencies I realised that the free data had a very wide range with some extreme outliers. As most of the data is already presented in categories (ie 1, 2, 3 etc) I took the decision to convert all the data into this form. This is now done but again valuable time has been used.
Now that I only have data in categories the next frustration was realising that simple ANOVA is not suited to this as there are no figures to provide means. So which statistic to use? Rather than follow trial and error I have spent over a week re-reading the statistics books to refresh the bonce! What I am interested in is the relationship between variables to see where further investigation is needed. Correlation will provide this, but thereby results another frustration - there are different methods of determining this in SPSS! The method most suited to my categorised data is Spearman's rho correlation. I have now started to put the data through this but the SPSS output is immense. Consequently I am taking it into Excel so that I can both process and sort it but also to condense it for printing out! It may be that I will convert some data back to values at some point but at the moment I have plenty to do.
That is the progress so what is the frustration. Needless to say that is the statistics. My intention of using ANOVA was scuppered once I started to look more closely at the detail. Originally I had a mix of free and categorised data. When I looked again at the frequencies I realised that the free data had a very wide range with some extreme outliers. As most of the data is already presented in categories (ie 1, 2, 3 etc) I took the decision to convert all the data into this form. This is now done but again valuable time has been used.
Now that I only have data in categories the next frustration was realising that simple ANOVA is not suited to this as there are no figures to provide means. So which statistic to use? Rather than follow trial and error I have spent over a week re-reading the statistics books to refresh the bonce! What I am interested in is the relationship between variables to see where further investigation is needed. Correlation will provide this, but thereby results another frustration - there are different methods of determining this in SPSS! The method most suited to my categorised data is Spearman's rho correlation. I have now started to put the data through this but the SPSS output is immense. Consequently I am taking it into Excel so that I can both process and sort it but also to condense it for printing out! It may be that I will convert some data back to values at some point but at the moment I have plenty to do.
Subscribe to:
Posts (Atom)
