Take the 2-minute tour ×
Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It's 100% free, no registration required.

I am currently carrying out an investigation to find if certain factors such as playing home or away or position of a footballer affects overall pass completion using logistic regression. I am using R to compute my data. In my current section in which I am trying to analyse uses the data of every player to convey a general conclusion to whether or not the position of a player affects the successfulness of pass completion.

so far I have computed:

test.logit <- glm( cbind(Total.Successful.Passes.All,Total.Unsuccessful.Passes.All) ~
                   as.factor(Position.Id), data=passes.data, family = "binomial")

summary(test.logit)

and my output was:

Coefficients:
                        Estimate Std. Error z value Pr(>|z|)    

(Intercept)              0.28482    0.01256   22.67   <2e-16 

as.factor(Position.Id)2  0.99768    0.01438   69.38   <2e-16 

as.factor(Position.Id)4  1.06679    0.01398   76.29   <2e-16 

as.factor(Position.Id)6  0.68090    0.01652   41.23   <2e-16 

Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 32638  on 10269  degrees of freedom
Residual deviance: 26499  on 10266  degrees of freedom

AIC: 60422

Number of Fisher Scoring iterations: 4

the intercept is goalkeepers,position.Id 2 is for a defender, 4 = midfielder and 6 = striker

Is this a good set of results to come to a conclusion? and with the large deviances?

share|improve this question

migrated from stackoverflow.com Jan 15 '13 at 18:36

This question came from our site for professional and enthusiast programmers.

2  
This doesn't feel like a programming question, but rather one about how to interpret the results of a logistic regression. As such, I think this is off topic for SO. I would suggest consulting a statistics textbook, or even better a statistician. –  joran Jan 15 '13 at 15:58

1 Answer 1

up vote 1 down vote accepted

Based on

  1. Very small p-values (less than 2e-16)
  2. Large effect size estimations (log-odds ratios ranging from .68 to 1.06)
  3. Small standard errors (around .015)

All evidence indicates there is a significant effect of position on the probability of a successful pass.

Note that the size of the deviance isn't immediately interpretable as "large" or "small" (deviance is affected by the sample size, so there is no such thing as a "large deviance" without context).

share|improve this answer
1  
Position.Id here is a categorical variable. Therefore, doesn't the estimate and p-values show significance only with relation to the reference (which is not shown in the post) (what I mean is that if the factors were re-levelled, you might get different p-values)? Also, how important is the ratio of residual df to residual deviance in a logistic regression? Could you please comment on it? Thank you! –  Arun Jan 15 '13 at 16:11
    
thank you very much for your help, it is much appreciated –  user54511 Jan 15 '13 at 16:23
1  
@Arun: Yes, they do show significance relative only to the reference level. Note that 2 and 4 are therefore very close together (though probably still far apart enough, given those standard deviations, to be statistically significant). The OP should try releveling the factors to check that. The residual deviance and null deviance mostly play into the question of how much of the variance this regressor explains (while there is no equivalent to R^2 in logistic regression, there are some "pseudo-R^2"'s that use these deviances. –  David Robinson Jan 15 '13 at 16:49

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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