I am a university level psychologists-come-English-conversation teacher, that

directs student psychology even though I don't have much of a clue about stats.

Generally I get my undergrad students to do a simple correlation research survey

to test a hypothesis. We also use t-tests in two conditions, and occasionally chi

squared tests when the response is a choice.

One of my students wanted to test the hypothesis that "our underwear represents

our personality more than our outer garments". She is really into underwear.

So we had subjects rate their personality, their underwear, their outer garments,

and their ego ideal in a number of dimensions. Brainy, even tempered, sociable etc.

In order to test whether which of these four matched the most I did two things.

1) Calculated the difference between them (using =ABS(....)) and performed t-tests

on the average difference. (The data did seem to show that underwear is the best

judge/expression of character. )

2) Calculated the correlation coefficient within subjects, in attempt to see if their

rating of their underwear, outer garments correlate with their rating of their ego

and ego ideal (using =Correl(.....)) and then averaged these correlations across

the students to find e.g. the average correlation between clothing ratings and

personality ratings for the subjects. (This time outer garments showed higher

average correlation with ego/ego-ideal)

It is this second attempt that I would like to ask about. Is this an acceptable

way of testing the similarity between the two sets of ratings?

Embarrassed, and grateful you have read this far!

Timothy Takemoto

Yamaguchi University ]]>

I am trying understand how to correctly build a mixed-effects logistic regression model in R. I believe my model is pretty simple and straight forward but I'm lacking in experience and uncertain I'm doing it correctly. Not being a statistician and struggling with stats, the specifics evade me.

Data: I have a word discrimination task where participants hear a word then must choose between two images on a screen. This task was run with 30 speakers from three language dialects: Australia English, UK English, and US English.

Research question: Is there a significant difference among the languages across the continuum as a whole e.g., Australia-UK, Australia-US, UK-US? Image one (if it appeared) shows the trend lines in the data. I am only interested if the general trends e.g., is UK perception different form US perception. I am not looking for significant differences at each point along the continuum e.g., at point 3 is there a sig. diff. between the language pairs?

Dependent variable:

Fixed effects:

Random effects:

I believe I will need three models with the following subsets of data:

Data set for model 1: Australia-UK subset Data set for model 2: Australia-US subset Data set for model 3: UK-US subset

I am also confused with the following parameters:

- Should I be using: glmer or lmer?

- Do I need family=binomial?

- Does continuum need to be alpha? It's currently numeric.

- Do I need an interaction? e.g., language * continuum

Attempted models and a result from the first:

Code:

`Australia_UK=lmer(Response~Language+continuum+(1|Speaker)+(1|Word), data=Aust_UK)`

Australia_UK=glmer(Response~Language+continuum+(1|Speaker)+(1|Word), family=binomial, data=Aust_UK)

Australia_UK=glmer(Response~Language*continuum+(1|Speaker)+(1|Word), family=binomial, data=Aust_UK)

Australia_UK=lmer(Response~Language+contiuum+(1|Speaker)+(1|Word), family=binomial, data=Aust_UK)

Linear mixed model fit by REML

Formula: Response ~ Language + continuum + (1 | Speaker) + (1 | Word)

Data: PercVDB_ie_ML_Q

AIC BIC logLik deviance REMLdev

587.1 617.2 -287.5 558.8 575.1

Random effects:

Groups Name Variance Std.Dev.

Speaker (Intercept) 0.024772 0.15739

Word (Intercept) 0.010677 0.10333

Residual 0.090904 0.30150

Number of obs: 1126, groups: Speaker, 30; Word, 2

Fixed effects:

Estimate Std. Error t value

(Intercept) 0.596684 0.082164 7.262

LanguageUK -0.003655 0.071121 -0.051

continuum -0.060863 0.003793 -16.047

Correlation of Fixed Effects:

(Intr) LnggUK

LanguageUK -0.409

continuum -0.342 -0.003

I'm an undergraduate student and I've designed an experiment for one of my capstone courses, but I'm not sure quite how to analyze the results.

I'm having my participants play fixed iterated games of the prisoner's dilemma. There are 6 participants in a session and they each play every other participant once, and then they play three rematches which are randomly chosen. Against each player they play 5 rounds/games of the prisoner's dilemma. They know that they will play exactly five games with each other player.

My independent variable is the amount/type of pre-play communication they are allowed to engage in with three levels:

1) full communication- players are allowed to communicate by speaking for 45 seconds before deciding whether to cooperate or defect

2) limited communication- players are allowed to communicate only with gesture, facial expression, or body language for 45 seconds before deciding whether to defect or cooperate

3) no communication- players are not allowed to communicate at all during the 45 seconds before deciding whether to defect or cooperate

My dependent variable is "cooperation" or "cooperative outcomes"

I'm getting ALOT more cooperate outcomes (cooperate/cooperate) in the full communication condition, as expected. I just don't know how to enter this into SPSS or what test to use. I thought about chi-squared originally, but not I'm unsure. Can anyone help me? ]]>