4
$\begingroup$

I am currently completing my dissertation. My study is cross-cultural and looks at predictors and inhibitors to adoption of technology in two countries (Thailand and Australia). I have a hypothesised model with IV's (Ease of Use, Usefulness, Need for Interaction, Risk, and Social Influence) directly linked to a single DV (intention to use). Both models are the exact same, so are the IVs and DVs (and related items), and sample sizes are similiar.

I have run regression analysis on both the Thai and Australian sample individually. I have the regression coefficient outputs with signifiance etc. What I am trying to find out now is how to best test the following question (or something similiar): "Social influence (IV) will have a stronger relationship in Thailand with intention to use (DV) m-banking than in Australia".

Is this the best way to test whether individual constructs fit better in one country then another? I want to test each individual construct to find out which has a more significant relationship between that IV and DV.

I apologise if this question has been answered already somewhere on the site or sounds very simplistic. I am using SPSS v19.0 btw. Thanks in advance!

Improve this question
$\endgroup$
4
  • 1
    $\begingroup$ Do you really want to test significance? That is as much a measure of how much data you have as it is a characterization of the data. The questions of scientific interest would focus on effect sizes (that is, the coefficients) and goodness of fit diagnostics (such as tests for interactions and nonlinear relations). BTW, welcome to our site! It's nice to see such a thoughtful, important, well-expressed question. $\endgroup$
    – whuber ♦
    Commented Oct 11, 2012 at 18:29
  • 1
    $\begingroup$ The key concept ( / keyword to search for) here is that of an interaction. Having a dummy variable coding for Thai vs. Australian will let you know if one group tends to have a higher intention to use on average; forming an interaction between that dummy and a covariate of interest will let you know if the relationship between the other IV and the DV differs b/t the groups. $\endgroup$
    – gung - Reinstate Monica
    Commented Oct 11, 2012 at 18:29
  • $\begingroup$ Hi Whuber/Gung - I just saw these responses! Looking at the standardized coefficients would make sense. Again, apologies for the ignorance but how would one test the difference in effect sizes. I conducted a CFA and found GFI for the model was similiar. Not sure if that is relevant $\endgroup$
    – Ben
    Commented Oct 11, 2012 at 18:37
  • $\begingroup$ Ben, since the estimates are based on independent data sets, you can consider two comparable effects each to have an independent sampling distribution (which is estimable from the data). Simply compare those two distributions. For instance, if the sampling distribution is approximately normal in each case, just do a Z-test based on the estimated effect sizes and their standard errors. A little more care may be needed when testing many pairs of effects, but all the information you need will often be present in the estimates and their covariance matrices. $\endgroup$
    – whuber ♦
    Commented Oct 11, 2012 at 21:32

2 Answers 2

Reset to default
3
$\begingroup$

The best way to test this is to combine the two samples, then add a variable for country and then test the interaction between the other IVs and country. This gives you everything you would get for an ordinary regression - effect sizes, standard errors, p values etc. for the interaction you want to test.

Improve this answer
$\endgroup$
7
  • $\begingroup$ Thanks for the answer Peter. Perhaps you could expand a little bit on how I would go about this? Are you suggesting I use linear regression - if so, where would the country variable fit? Excuse my ignorance. $\endgroup$
    – Ben
    Commented Oct 11, 2012 at 18:15
  • $\begingroup$ I don't use SPSS so I don't know how to do it in that program, but the variable "fits" in the same place as all the other variables. It could be coded A and T. how did you get your data into SPSS? $\endgroup$
    – Peter Flom
    Commented Oct 11, 2012 at 18:22
  • $\begingroup$ It was imported from keysurvey, a online survey tool. I have a country variable. I am just unsure how including a country variable in a linear regression (along with other IVs) would allow me to test the question "Social influence (IV) will have a stronger relationship in Thailand with intention to use (DV) m-banking than in Australia". $\endgroup$
    – Ben
    Commented Oct 11, 2012 at 18:23
  • $\begingroup$ That is tested by the interaction between country and social influence. $\endgroup$
    – Peter Flom
    Commented Oct 11, 2012 at 18:24
  • 1
    $\begingroup$ whuber makes an important point about it being subtle - studying an interaction within the same study raises less concerns about that implicit assumption of common error/lack of fit. It is good that you have the same IVs in both, but were there measured with the same accuracy over the same range? This type of extra concern about heterogeneity may justify thinking of it as a meta-analysis. Your situation is briefly addressed in Greenland S, O' Rourke K: Meta-Analysis. Page 652 in Modern Epidemiology, 3rd ed. Edited by Rothman KJ, Greenland S, Lash T. Lippincott Williams and Wilkins; 2008 $\endgroup$
    – phaneron
    Commented Oct 11, 2012 at 19:57
1
$\begingroup$

@whuber makes an important point about it being subtle that I commented on but wish to put in as an answer.

It’s understandable that one would think of two studies as being like two strata within one study. But as David Cox once pointed out, there are much greater concerns about homogeneity or what can be taken as common. For instance, the within group variance is often taken to be the same in different factors but this is seldom true between two different studies.

It is good that you have the same IVs in both, but were there measured with the same accuracy over the same range? This type of extra concern about heterogeneity error/lack of fit may justify thinking of it as a meta-analysis.

Your situation is briefly addressed in Greenland S, O' Rourke K: Meta-Analysis. Page 652 in Modern Epidemiology, 3rd ed. Edited by Rothman KJ, Greenland S, Lash T. Lippincott Williams and Wilkins; 2008

Improve this answer
$\endgroup$

Your Answer

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

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