# Help to choose appropriate statistics for research

#### mag

##### New Member
I am working on my first research project. I am really junior in this matter. My research aims to explore factors associated with change of financial behaviours among over indebted consumers.

I have to make some decisions about which statistical analyses are appropriate for my sample size and for my research questions.

Is there anybody in this forum that I can contact with a private message? Or should I post the summary of my research here?

Thank you,
Mag

#### JohnM

##### TS Contributor
mag,

post a summary here.

research rule #1 - make sure your research questions are well thought out and articulated long before you begin thinking about stats methods and sample sizes.

#### mag

##### New Member
The objective
My project aims to explore factors associated with adoption of positive financial behaviour by clients of a financial counselling agency.

The Theoretical Framework
I am using the Transtheoretical Model of Change (TTM) framework. The TTM states that people change behaviours by stages and that in each stage they use different processes. According to the TTM the processes of change are the stronger predictors of behavioural change.

The hypotheses
Following the model I put together a number of hypotheses:
• H1. Previous experience with positive financial behaviour is a predictor of more consistent practice of the behaviours.
• H2. The more helpful aspects perceived for the counselling agency, the higher the probability of adopting positive financial behaviours.
• H3. The more processes of change experienced the higher the probability of consistent practice of positive financial behaviours.
• H4. The more education achieved, the more likely the client is to act upon adopting positive financial behaviours
• H5. The older the client, the more likely he or she is to act upon adopting positive financial behaviours
• H6. The higher the income, the more likely the client is to act upon adopting positive financial behaviours
• H7. Respondents who acknowledge responsibility for personal financial situation are more likely to implement positive financial behaviours

Sample
So far I know that my sample will be small. The agency will look into the files of each person who contacted the agency between January 1st and March 31st 2007, and enlist those who have agreed to be contacted to provide feedback. We will contact all of them and hope to have at least 60 respondents.

Questions to the consultant
1. What are the statistical analyses that I should use to answer my research questions/hypotheses?
2. What statistical analyses are appropriate for my sample size and for my research questions?
3. Would you have any suggestions on how to proceed with the analysis?

I have some ideas of what to do. I now I can use Chi Square to measure strength of association after doing crosstabulation. I also understand that regression analysis would not be apropriated for a small sample when using so many variables. I need guidance to decide what to do.

I adapted a survey from previews study to use on my research. It can be seing at https://www.surveymonkey.com/s.aspx?sm=lzQ6os_2f46fstQNFQkKH6cg_3d_3d

A study using similar instruments and framework can be found in the following link:
http://www.afcpe.org/doc/Vol1529.pdf

#### JohnM

##### TS Contributor
Well done! What I would do next is figure out how you intend to measure the independent and dependent variables in H1 - H7 - i.e., categories, numerical score, etc.

Then we'll be able to start looking at statistical analysis methods.

#### mag

##### New Member
Okay. The measures is something I don't understand very well. I put my questions on Survey Monkey. I wonder if you cna help me to understand what exactly "how I will measure" means by looking at my questions.

To test my first hypothesis I will ask if individuals practice a series of financial behaviours. According to their responses I will identify in which stages of change they are and whether they started the behaviour after contacting the counselling agency. I am predicting that if they have started the behaviour before contacting the agency they are more likely to practice the behaviour more consistently.

My next question asks how often they practice five financial behaviours. They will respond in a scale from one to five - from never to repeatedly.

Than I will ask them to rate the helpfulness of diffent aspects fo the agency they contacted. They will answer in a scale from one to five - from not helpful to extremelly helpful.

Than I will ask about processes they may have used that may have helped them to adopt the positive behaviours. They will answer in a scale from one to five - from never to repeatedly.

Can you help me to understand the measures?

Last edited:

#### jkotlerman

##### New Member
you have a lot of different variables, but from what I have read here, it seems that you are looking at factors that increase positive financial behaviors. A lot of your variables are on a likert scale, are you planning to measure each behavior separately or combining them into one?
Before you start to figure out which statistical analysis you need to use, you need to figure out what are your outcome variables and what are your predictor variables. Predictor variables are those independent variables that predict the outcome of the dependent variable.

Jenny Kotlerman
www.statisticalconsultingnetwork.com

#### mag

##### New Member
My dependent variable is the adoption of positive financial behaviours. I am asking about 13 positive financial behaviours do know whether or not people practice those behaviours. Than I summarize those behaviours into five positive financial behaviours and ask how often people practice them.They will respond with a likert scale.

My predictor variables are:
1 Previous experience with positive financial behaviour (TTM scale & number of financial behaviours practiced -13 behaviours)
2 Perception of agency helpfulness (likert scale & number of help aspects perceived - 22 aspects)
3 Processes of change experienced (likert scale & number of processes of change - 20 processes)
4 Educational achievement (from lower to higher - 5 options)
5 Age ( I am aking for a number, but think should ask for a range)
6 Income (from lower to higher - 11 options adding $1,000 from$0 to \$10,001)
7 Perceived reason for ifnancial crisis (19 reasons - 5 of them indicate personal responsibility for the situation - those are the one I predict will take action to adopt the positive financial behaviours)

The direction of the relationship is as follow:

1. Previous experience (answer number 5 in the scale - "yes, I have being practicing this behaviours for more than 6 months") increase likelihood of adopting positive financial behaviours

2. higher number of helpful aspects perceived increase likelihood of adopting positive financial behaviours

3. higher number of processes of change experienced increase likelihood of adopting positive financial behaviours

4. higher education increase likelihood of adopting positive financial behaviours

5. older age increase likelihood of adopting positive financial behaviours

6. higher income increase likelihood of adopting positive financial behaviours

7. acknowledgin mismanagement as reason for financial crisis increase likelihood of adopting positive financial behaviours

Does it make sense?
I am not sure the scores are done. Can ou send me a hint on that too?

Thanks,

#### JohnM

##### TS Contributor
Jenny,

My impression is that the purpose of this study is to identify which of the 7 hypothesized factors are good predictors of positive behavior, which would then lead to further research, maybe into building a model for predicting positive behavior. I don't think this research is at the model-building stage, more on factor testing and selection.

In terms of indep var's and dep vars, I think she's pretty much settled that one - my take:

Indep variables, watered-down a bit:
(1) previous experience w/positive behaviors
(3) previous experience with processes of change
(4) education
(5) income
(6) taking personal responsibility

Dep variable:
(1) likelihood (or degree) of adopting positive financial behavior

Mag,
Since most, if not all of your variables are Likert scales or something similar, you could try regression or correlation - since Likert scales, strictly speaking, are ordinal, a nonparametric approach such as Spearman's rho would work (very similar to the Pearson correlation coefficient, but for ranked data). However, there isn't any harm in using Pearson's r as well to see what you get.

There are tons of resources online for conducting and interpreting correlations.

John

#### mag

##### New Member
combining behaviours

"are you planning to measure each behavior separately or combining them into one? "

I guess what I have to do is to combine all the optins in the questions as one, but I am not sure how it work.
What are my options? What are the advantages of combining in one? How do I do it?

Thanks,

#### JohnM

##### TS Contributor
I don't think your study is ready for that yet - just analyze them separately for now....(see my previous post above)

#### mag

##### New Member
Okay. Thank you!

You make it sound so simple! You have the clarity I am looking foward to develop. I am sure after all this learning process (and struggle) I will be able to simplify things too.

I have done regression and correlation before. And I have used pearson's r too. I will have to study a little bit since I don't remember how to do it. I will seek those resources online too.

Thank you again. I will sure post again with more questions.
Cheers,
Mag

#### jkotlerman

##### New Member
I agree that your study is not ready for either the analysis stage nor creating new combination variables, but if you are writing a plan for the study, then you need all potential variables (including the ones that you might create in the future) and analyses listed before you start analyzing.
So for your future analysis I still have a few questions.
1.Is your outcome variable Having positive financial behavior (yes/no) or some scale of degree of financial behavior?
2.Also in your list of predictor variables, steps 1,2 and 3 mention behaviors, aspects and processes. Are these individual variables or one variable representing the number of each?

For now,if you are eager to start your analysis, you need to calculate your spearman correlations first, just like John said, to look at the associations between your predictors. Also you should look at the crosstabs.

Also, leave your age question as an open question, it is always easier to combine into categories later.

Jenny Kotlerman
www.statisticalconsultingnetwork.com

#### JohnM

##### TS Contributor
".....but if you are writing a plan for the study, then you need all potential variables (including the ones that you might create in the future) and analyses listed before you start analyzing. "

I think you would run the risk of getting ahead of yourself and not focus on the specific objective of the current study. I would recommend holding off on creating / discussing future variables until you start writing the discussion of the results (i.e., if you notice an interesting pattern or trend or if it looks like 2 or more independent variables are measuring basically the same thing), and the section at the end which talks about potential directions for future research.

I know it is sometimes tempting to change / adjust the direction of research based on initial results, but (and this is just from my own experience) I would wait until this study is done and these hypotheses have been tested.

I agree that looking at cross-tabs in addition to using spearman's rho would be beneficial, because the sample size for the study may be small, and Likert scales sometimes are "resistant" to statistical significance. What I mean here is that it is sometimes difficult to obtain significance with a small scale (1-5).