I am carrying out my dissertation research on how autobiographical referencing and different advertising medium on consumers' past memory.

I am encounter a major problem. All participants took a test called the Life events inventory to rate how certain they were that an event had happened to them as a child. This was measured in a likert scale. The critical item on the LEI appeared as a fourth item 'met and shook hands with a famous tv character at a theme resort.'

A week later, 1/2 participants were exposed to an autobiographical Disney ad which focused on a child's experience of shaking hands with mickey while the non-autobiographical Disney ad focused on future visits with a dramatic account of all the major landmarks of the theme park. The ads were presented in either written, audio and written and audio form. Afterwards, they were asked to complete the LEI again.

The problem is that I don't know which stat test to use. I could use chi2 but this isn't a very powerful test. The dependent varible is the difference between the LEI scores between week one and week two. I could use an anova which requires interval/ratio data. but I have to mention the debate about the what is ordinal data?(my dependent varible is nominal/ categorial)

Someone suggested using of log-linear modelling. I have done some reading but still confused as to what it really is?

-- maggie lo (lo_maggie@hotmail.com), February 12, 2003

Answers

Hi Maggie, it may be too late this go around for you, but a distinguished professor I worked for taught me this basic rule of research: never collect any data if you don't know how you are going to analyze it. Now you have to resort to what some skeptics in this business call snooping your data. You might forward your question to dand@yorku.ca. Best, David

-- david clark (doclark@yorku.ca), February 14, 2003.

thanks for your advice.

The problem was I misread the original study carried out by Loftus. They asked participants to rate their certainty on a 100mm, whereas I took it as any scale would perform the same job. I was hoping to use a t test to analysis. Nevermind, I have found something about the log linear modelling on the net that might be useful.

-- maggie lo (lo_maggie@hotmail.com), February 18, 2003.