Please please please help me

#1
Masters Dissertation Data Analysis - Linear Ordinal Regression

Hi, I'm really hoping that someone on here can help me. I'm sure this is going to be something ridiculously simple for someone but I am struggling big time! I'm currently doing my masters dissertation looking at whether there is any link between coaching behaviours exhibited by PE teachers (independent), and the motivation levels of their pupils (dependent). Both variables are assessed by answering a questionnaire with likert values as the answers (making it ordinal?) Please correct me if I go wrong at any time.

I think the first bit is simple.... I have run a bivariate correlation analysis (spearmans, not pearsons) and gained my correlations between variables. This bit I think I'm OK with. Secondly, I want to work out whether one variable can predict another. I'm thinking that to do this, I need to run either a linear or ordinal regression analysis. THis is where I get beaten as I put the numbers in, and the output is complete gobbledegook! Is it necessary to do one, am I going to get any decent data from it, and if so, what do I need to look at?

Thanks in advance

Nick
 
Last edited:

hlsmith

Omega Contributor
#2
What is your sample size and how many independent variables will there potentially be and how will they be formatted?
 

noetsi

Fortran must die
#3
Secondly, I want to work out whether one variable can predict another. I'm thinking that to do this, I need to run either a linear or ordinal regression analysis. THis is where I get beaten as I put the numbers in, and the output is complete gobbledegook! Is it necessary to do one, am I going to get any decent data from it, and if so, what do I need to look at?
If your scale of the dependent variable is likert (particularly if you have less than 7-9 unique levels) than ordinal logistic regression is strongly recommended if you are going to do regression. Formally I believe such an ordinal dependent variable should never be used with linear regression, but in practice many feel if you have enough different levels than you can treat it as "interval like" and thus use linear regression. Also as the number of unique levels gets greater it becomes more difficult to interpret ordinal logistic regression.

WHen you say you output is "gobbledgook" what do you mean specifically? There are many ways this could occur, we need to see your output or have you describe in detail what the problems are.
 
#4
Right, well the first questionnaire is based on coaching behaviours, and is 36 questions answered always - never (5 points). Whereas the other questionnaire is based on motivation, and is 45 questions answered from strongly disagree - strongly agree (7 points). I'm sure the data I'm getting is OK, but I have no idea how to interpret it.

I want to see whether motivation levels can be predicted by coaching behaviours. I'm sure its a simple test of stats, but my knowledge is shaky to say the least. I could show it to you but I would have no idea how to upload it.
 
#5
How did that questionnaire come together? It sounds to me that you used a survey instrument that was academically-designed and validated at some point in the past. Is that correct?

If so, I assume that you first need to set up a latent construct that measures motivation and coaching based on those sets of questions. In other words, this has all the makings of a structural equation modeling to me.
 
#7
Do you know if the model you're trying to create now was created back then as well? Did you replicate what was done, and the output is completely different than the previous time's output?

And generally, what do you mean by gobbledygook? Is it an issue that you don't know how to interpret the output, or that the output indicates absolutely no interaction between the constructs you're measuring?
 

hlsmith

Omega Contributor
#8
What is your sample size and how many independent variables will there potentially be and how will they be formatted?
Are you going to answer these questions or just skip them? It is unclear if you just want to compare two questions off two different questionnaires or if you are looking at established constructs from questionnaires and even if the same people completed both questionnaires?
 

Karabiner

TS Contributor
#9
please please me oh yea like I please you
Help, I need somebody, Help, not just anybody
Help, you know I need someone, help

(yesterday...?)
 
#10
Are you going to answer these questions or just skip them? It is unclear if you just want to compare two questions off two different questionnaires or if you are looking at established constructs from questionnaires and even if the same people completed both questionnaires?
Sorry I missed this question. I gave the same 2 questionnaires to 420 pupils in years 8-11. I'm looking to compare the 2 questionnaires
 
#11
I'll try and give you a breakdown of what I've done but apologies if I miss something. I asked 420 pupils to rate the coaching behaviours exhibited by their teachers (using the LSS), then the pupils levels of motivation (using the BRSQ). The LSS gives the mean coaching behaviours across 5 criteria (Training and Instruction, Democratic behaviour, autocratic behaviour, social support, and positive feedback). Anything higher than 2.5 means a behaviour exhibited often. The BRSQ judges motivation across 7 criteria (1-7), and higher than 3.5 means they have that direction of motivation.

So basically I have done a spearmans correlation between the outputs of the first questionnaire and the outputs of the second, looking at whether high levels of coaching behaviour correlate to high levels of motivation, low levels etc. Now I want to see whether a) high levels of exhibited coaching behaviour can be predictors of high levels of motivation (using each of the 7 motivation strands as dependent variables(?)), and b) how much of the motivation levels can be related to coaching behaviours (maybe a nagelkierke square?)

Any better?
 

Karabiner

TS Contributor
#12
Seemingly you have scales which consist of several Likert-type
items. You probably can treat these scales as interval.

to see whether a) high levels of exhibited coaching behaviour can be predictors of high levels of motivation (using each of the 7 motivation strands as dependent variables
has already been done (bivariate correlations), for answering
how much of the motivation levels can be related to coaching behaviours
you could try a multiple linear regression.

With kind regards

K.
 
#13
Thank you so much. I appreciate your help, although when I run the analysis, how can I interpret the data.

Am I looking at the model summary, anova, or coefficients output tables? What means what?

Thanks again
 

Karabiner

TS Contributor
#14
I see. So maybe you better abstain from anything more complicated than
correlation and do a little reading first. E.g. Andy Field's "Discovering
Statistics Using SPSS" has been described as a very helpful introduction
for beginners. Also, there are websites which explain the output of
statistsical software in detail (at least I suppose so).

With kind regards

K.
 

noetsi

Fortran must die
#15
Generally speaking if you have a dependent variable that is a collection of a series of likert scale questions you can treat it as interval.