1
$\begingroup$

I wish to analyse the following :

Independent Variable (IV): User of online financial reports Perceived usefulness and Perceived Quality (Sub to Relevance, Reliability, Understandability, Comparability and Timeliness) (all using 5-point Likert scale.) towards Internet Financial Reporting in Malaysia.

Dependent Variable (DV): Intention to use Internet Financial Reporting information as a decision-making instrument (5-point Likert scale)

Questions:

  1. Which statistical method should I use to analyse my data? Previously I was thinking to use multiple regression and correlation to analyse it, but I just found that I'm not fulfilling the assumptions, because the DV should be measured on a continuous scale.

  2. What are the minimum numbers of observations I need? I read before that it is at least 10 for each IV. Does that mean I only need 60 observations?

I'm using SPSS to analyse the data.

$\endgroup$
1
  • $\begingroup$ It's worth emphasizing something @Nick said: "As all your predictors are also categorical, you will be estimating a lot of parameters". If you're treating e.g. Understandability as a continuous variable, it's one parameter for a linear effect, two for a quadratic, &c.; if you're treating it as categorical it's four parameters (no. categories excluding a reference level). Interactions between IVs add more parameters. Any rules of thumb should be talking about the no. parameters, not the no. IVs. And it's sensible to consider the size of effect you want to be able to estimate. $\endgroup$ May 22, 2013 at 11:15

1 Answer 1

3
$\begingroup$

I don't think continuous scale is absolutely essential for regression of the usual kind. There are plenty of examples in which I might use regression on a response (what you call DV) that is a counted variable.

But in essence I think you are right. The key word here is ordinal or ordered: your response variable (DV in your notation) is on an ordinal or ordered scale. The books of Alan Agresti are among those relevant.

http://www.amazon.com/Analysis-Ordinal-Categorical-Probability-Statistics/dp/0470082895/

gives an in-depth survey and his other books on categorical data analysis are also excellent. And there are many other surveys.

Any rule of thumb "at least 10 for each IV" implies a minimum, not to be thought of "what will necessarily work well". As all your predictors are also categorical, you will be estimating a lot of parameters and 60 to me looks very small for such a project.

Probably the first model to look at is ordered logit, but there many different models in this territory.

It's just terminology, but a widely held view is that dependent variable and independent variable are poor terms and hang-overs from the past. That said, there is no agreement on what to use instead, but response and outcome are popular terms rather than dependent variable and predictors and covariates are among several popular terms rather than independent variables. If you use IV as an abbreviation, you will get into confusing discussions with those people who want to use IV to mean instrumental variable.

$\endgroup$
8
  • 1
    $\begingroup$ Re your last paragraph, I think this could make an interesting community discussion. I kind of like DV and IV (and, so far, I've not had anyone get confused and think it was an instrumental variable). Perhaps this varies by field? $\endgroup$
    – Peter Flom
    May 22, 2013 at 10:22
  • $\begingroup$ I kind of hate DV and IV. I was brought up with the explanation "Deo volente" for DV. The deeper objections are (a) easy for newcomers to mix them up (b) dependence and independence are overloaded terms in probability and statistics (c) good, crisp, evocative alternatives are available. I teach students to talk of responses and predictors, for example. Economists and people influenced by econometrics (e.g. in sociology and political science) often use IV for instrumental variable; as many statistical fields never use instrumental variables, it is easy not to come across this usage. $\endgroup$
    – Nick Cox
    May 22, 2013 at 10:32
  • 1
    $\begingroup$ I always thought of (in)dependent [variables] as the only appropriate terminology when talking about experiments (which is not the context of this particular question), not as something of the past. Other terms (response/outcome/predictors) are simply unknown in some fields. “Predictor” in particular does not feel really appropriate when you are actually manipulating something. $\endgroup$
    – Gala
    May 22, 2013 at 10:33
  • 1
    $\begingroup$ The discussion that springs first to mind is Mosteller, F. and Tukey, J.W. 1977. Data analysis and regression. Reading, MA: Addison-Wesley. Most mainstream regression texts seem to avoid dependent and independent variables as terms. We're statistical people, but has any one good data on usage frequency? We all have our impressions and our own patchy coverage of the literature. Some psychologists talk about criterion variables; I have never understood that usage, but it seems popular. $\endgroup$
    – Nick Cox
    May 22, 2013 at 10:37
  • 1
    $\begingroup$ @GaëlLaurans: I agree. Experimentalists often need or prefer a language that matches how they talk about their experiments. Some of them talk about factors; they are usually not people who also do factor analysis! People who prefer independent variables as terminology have to explain that independent variables can be highly correlated with each other, perhaps one of the smaller problems we have in teaching. $\endgroup$
    – Nick Cox
    May 22, 2013 at 10:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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