Likert analysis for dummies?

[morphemass]

I find myself having to admit that I’m terrible with stats, embarrassingly so, but I need to use them within my thesis. I have absolutely zero confidence in my understanding or abilities when asked to use stats (15 years since I did my bachelors) hence I am laying out my problem and what I understand in the hope that if I have gone way off course someone will pull me up. As a bit of additional background – this is a Masters in a HCI related area, my supervisor has taken ill, hence (in part) why I’m asking here. I’ve two “styles” of gestural interaction which I am comparing, one which I call “discreet” the other “exaggerated”.

I have the following hypothesis : “There will be no difference in the social acceptability of the discreet gestural interaction style in comparison to the ‘exaggerated’ gestural interaction style”

I created a questionnaire (http://questionpro.com/t/AHanjZLzMW) containing 8 videos, 4 of which had one style (exaggerated), 4 of which had another style (discreet); randomised the presentation of each video and then had people rate on a Likert-type scale whether they would feel self-conscious or comfortable performing the gesture (3 levels of comfort vs 3 of self-consciousness). Muddying the waters somewhat I had them also apply the rating across location & audience so for each video I would have 5 ratings. I provide the link to clarify this explanation.

From my basic understanding, I am using my independent variable (gesture style) to test if there is a relationship to, as a measure of social acceptability, perceived self-consciousness/comfort. I am using the location/social situations as controlled variables since these are likely to influence the dependent variable.

The basic approach I had been thinking of was that if I tabulate & combine the results from the 4 exaggerated styles, and the results from the 4 discrete styles; I then have 2 tables containing counts of number of response for each item on the Likert-type scale, which I can then use to compare the 2 gesture styles. How valid is this approach and if valid, is there an optimal statistical test to use here?

I’ve tried to give sufficient background to my problem and what I’m trying to accomplish in the hope that someone can help me work though this; nods in the correct general direction with the occasional slap would be appreciated!

Many thanks in advance!

[morphemass]

Okay so no one has slapped me yet, does this mean that my question makes no sense or that it makes sense, but my situation is so dire that I am beyond saving....or that I need to try and move myself towards a solution?

So I know the basics: I should describe my data using median or mode and use the interquartile range of an expression of variance. Fair enough. I can grasp that.

I can also vaguely grasp the arguments regarding the use of non-parametric rather than parametric tests but only just....I also see chi squared suggested as a tool to perform hypothesis testing for Likert-type ordinal data and think I can get my head around it.

Looking at my initial post, would performing a chi-square test for each of my controlled variables, across the two conditions be a valid way of making statistic inferences from this data? Or am I way off base?

Please, any pointers are greatly appreciated.

[noetsi]

My first question is if you last had stats 15 years ago, why are you using them in your disertation :)


I created a questionnaire (http://questionpro.com/t/AHanjZLzMW) containing 8 videos, 4 of which had one style (exaggerated), 4 of which had another style (discreet); randomised the presentation of each video and then had people rate on a Likert-type scale whether they would feel self-conscious or comfortable performing the gesture (3 levels of comfort vs 3 of self-consciousness). Muddying the waters somewhat I had them also apply the rating across location & audience so for each video I would have 5 ratings. I provide the link to clarify this explanation.

Did you have one likert scale with six levels(say the 3 levels of comfort then the 3 levels of self consiousness) or two (one of 3 levels for comfort and and one of 3 for self-consious). And are they ordered (so one level is above or below the other) or are they purely nominal level?


From my basic understanding, I am using my independent variable (gesture style) to test if there is a relationship to, as a measure of social acceptability, perceived self-consciousness/comfort. I am using the location/social situations as controlled variables since these are likely to influence the dependent variable.

Are the two, or possibly one, likert scales you mentioned above your measure of your dependent variable? How is your dependent variable measured (and how many dependent variables are there)?


The basic approach I had been thinking of was that if I tabulate & combine the results from the 4 exaggerated styles, and the results from the 4 discrete styles; I then have 2 tables containing counts of number of response for each item on the Likert-type scale, which I can then use to compare the 2 gesture styles. How valid is this approach and if valid, is there an optimal statistical test to use here?

It depends on your major. To me this seems like a descriptive statistics method (nothing is actually being calculated) and the majors I know of probably would not accept it. But your field could be entirely different. The answer to your question probably won't be found here (most are statistics types :) ). It will be found with your committee and a literature review.

What does the literature in your field say, that is what approaches do they use for similar issues, and what has the chair of your committee said about this.

Note you say thesis so I assume this is a PHD. If not, who has to approve this?

[morphemass]

My first question is if you last had stats 15 years ago, why are you using them in your disertation :)

A good question - this is a piece of exploratory research within which I developed something and want to provide some "evidence" that it addresses a perceived issue. While I can certainly argue that it does so, some basic statistic work would increase the legitimacy of my argument.


Did you have one likert scale with six levels(say the 3 levels of comfort then the 3 levels of self consiousness) or two (one of 3 levels for comfort and and one of 3 for self-consious). And are they ordered (so one level is above or below the other) or are they purely nominal level?

A single scale of 6 points as in your first example. Ordered so that the 3 levels of self-consciousness can be viewed as the negative condition, the 3 comfort levels as positives.


Are the two, or possibly one, likert scales you mentioned above your measure of your dependent variable? How is your dependent variable measured (and how many dependent variables are there)?

I'm attempting to measure social-acceptability for which there is no generally accepted measure (or rather, there isn't one that I could find in relation to gestures)...that is an interesting question though, wish I had time to pursue it :(


It depends on your major. To me this seems like a descriptive statistics method (nothing is actually being calculated) and the majors I know of probably would not accept it. But your field could be entirely different. The answer to your question probably won't be found here (most are statistics types :) ). It will be found with your committee and a literature review.


What does the literature in your field say, that is what approaches do they use for similar issues, and what has the chair of your committee said about this.

To be honest, my field is often quite lax with the stats although this isn't true across the board. I would have asked this question of my supervisor but he is currently on sick leave.


Note you say thesis so I assume this is a PHD. If not, who has to approve this?

Fortunately this is just a masters. As said, I could probably drop the stats entirely, I really am making a rod to beat myself with. Although I'm struggling, I think I can justify the chi squared approach above - I really only need a basic rather than rigorous test and just want to avoid obvious mistakes...