Take the 2-minute tour ×
Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It's 100% free, no registration required.

I hope you can help me because at the moment I am confused and helpless. I am final year PhD student, and my supervisors asked me to find out statistical methods other than exploratory (percentage of findings) to present my data.

Here are my research data where we want to present consumer attitude and reasons towards disposal, with individual responses (N=70) collected via a questionnaire with the following items:

  1. Product is functional? yes/no/not answered
  2. If it is working can you pinpoint the failure point? yes/no/not answered give the failure point
  3. If it is working, did you buy new product? yes/no/not answered
  4. If it is not working did you buy new products? yes/no/not answered?
  5. Are you aware of other options? yes/no/not answered

I want to know how can I measure reliability, and what other statistics analysis I can present: GLM, ANOVA , MANOVA, etc.

share|improve this question
    
What do you mean by "reliability"? What is / are your response variables? What are your explanatory / predictor variables? Do you have a hypothesis going in to this project that you want to test? Can you state the goals of this study? –  gung Nov 13 '12 at 21:25
    
reliability of formative scale, –  abmarie Nov 13 '12 at 22:51
    
reliability test of formative scale I underestand I can not used Cornbach alpha,the explanatory variables I beleive dependent (functionality and awareness). but i am not sure of response variable but i presume it is independent could you please give advise on this too. I do not have hypothesis to test, the aim of study the reason consumer discard their product from functionalit point of view and was to assess consumers attitude and knowledge towards disposal etc. –  abmarie Nov 13 '12 at 23:05
    
You still need to say a bit more to get a good answer. Do you have any other data, for instance on characteristics of the respondents? How did you expect your data to answer your research question(s) when you were planning the study? Most practical application of statistical methods involves hypothesis testing. Are you sure you don't have a hypothesis, e.g. something you expected to find which the data might support or refute? –  Stuart Nov 14 '12 at 10:49
    
Other than hypothesis testing some things you could do include examining correlations between answers using the phi coefficient and perhaps partial correlations too. I'm not sure what you mean by reliability of a formative scale. You seem to be thinking of a situation where you have a set of questions that, together, measure a single psychological construct. Do you see your 5 questions as together measuring a single aspect of the person's psychology? It seems unlikely but I don't know the area. –  Stuart Nov 14 '12 at 10:55
show 3 more comments

1 Answer

You first need to think through your research questions and whether there are specific hypotheses that could be tested using the data. (Ideally you would have these hypotheses before collecting data and they would influence your sampling strategy as well as the questions you ask. I think your supervisors should have helped you work through this instead of just telling you to use statistical methods but anyway...)

I know nothing about the subject, but to give examples for illustration, you might hypothesise that:

  1. Products bought longer ago are less likely to be seen by users as functional.
  2. Some makes and models are more likely to be seen as functional than others.
  3. Some makes and models of product continue to be seen as functional for longer than others.
  4. The probability of buying a new product depends on whether it is perceived as functional, the make and model, the time since buying the original product, and awareness of other options.

(These hypotheses seem very roughly to fit within your research questions but that's up to you to decide. They would normally be derived from theory and prior research in your subject area.)

These hypotheses would commonly be tested using logit or probit models, because those models are well suited to cases where we have a response (dependent) variable that can be either 1 or 0 and are trying to explain how the probability of it being a 1 varies with changes in other ('explanatory' or 'independent') variables. For hypothesis 1-3 the dependent variable is (the probability of) whether the use said the product was functional. For hypothesis 4 the dependent variable is (the probability of) whether the user bought a new product.

MANOVA or ordinary least squares (OLS) linear regression - the two are essentially very similar - would also work, but are arguably less appropriate when the dependent variable is binary and we are interested in probabilities.

Your sample may be too small for some types of hypothesis testing. For instance if there were 50 different make/model combinations in a sample of 70, you will not be able to say very much about a particular make/model.

Before running tests make sure to set up appropriate variables based on the questions. For instance, decide what to do with the 'not answered', which would usually be coded as 'missing' and ignored in statistical analysis, but that could potentially bias the results and you might see the non-answers as somehow informative in themselves. Make/model categories would probably be coded as a series of dummy (0 or 1) variables.

As I said in the comments, I don't think it makes sense to try and examine reliability (by which I think you mean internal consistency) in this case. Internal consistency is where you have a set of questions all aiming to measure some single construct, and that doesn't seem to be the case here.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.