Data Analysis Method: Binary Logistic Regression
Why? Logistic regression, unlike ordinary regression techniques, does not assume a linear relationship within each variable. This makes is particularly useful when it comes to analyzing categorical data.
By comparing each category within the variable in a logistic regression table, we are able to see what categories have statistically significant different results to the base category (the category that is used to compare the other categories to, in order to see if there is a significant difference within that variable for each - or any - of the categories).
Using a statistical significance level of p=0.05 (95% Confidence) throughout the data, the following results showed significance (i.e. had p-values of less thanor equal to 0.05, and the 95% confidence interval for the odds-ratio did not encompass one).
Once the "No Opinions" have been eliminated:
Age: the categories 16-21 and 45+ are different, so age has an effect on the opinion a person has - as you become older, you are less likely to support the continuation of the Miss World contest. There does appear to be a general trend, that, with a greater sample size, we may have been able to be more certain about the difference between the 16-21 and the 22-30 categories, and the between the 16-21 category and 31-45 year olds.
Ignoring, temporarily, the statistical significance levels, 22-30 year olds are approximately twice as likely to support the abolition of the contest.
31-45 year olds are 2.63 times, and 45+people are 5.88 times more likely to support the abolition of the contest as 16-21 year olds surveyed.
When we include the statistical significance level, and look at the 95% confidence intervals for the odds ratios (in comparasion to the 16-21 age category) we get:
22-30: between 0.72 and 6.66 times more likely to support abolition
31-45: between 0.70 and 10 times more likely to support abolition
45+: between 1.3 and 25 times more likley to support abolition (note that this does not encompass 1)
When taken on its own, the background of a person appears to have no significant effect on the decision made, but when the effect of the interaction between gender and background was considered, some interesting results became apparent.
While there was little evidence of a difference in the interaction between gender and background when subjects from a rural background were compared to those from an urban (city) background (controlling for gender - comparing females and males seperately), there is a statistically significant difference when it comes to those from an urban (provincial town) background. This isn't obvious in the initial logistic regression of the background variable because of the quite marked split between males and females in this category. So there is some evidence to suggest that provincial towns in Ireland are indeed a hotbed of feminism!
There is no statistically significant evidence to support a direct male/female divide, once the "no opinions" are removed, but this is more to do with the sample size and structure than anything else.
What about the "No Opinions"?
When we examined the differences between gender, age and background, when it came to whether the respondent had an opinio or not (grouping Yes and No categories together) the results were somewhat inconclusive. While there was some evidence that the 22-30 category was different to the 16-21 category, with a p-value of 0.042, the test that all slopes are zero had a p-value of 0.183, suggesting that, overall, we can make very limited differential statements between those with and without opinions.
If we were to be given the answer to the "Do you think that the Miss World contest should be abolished?" we would predict the following:
No Opinion Male with probability 0.638 and Female with probability 0.362
No Male with probability 0.652 and Female with probability 0.348
Yes Male with probability 0.367 and Female with probability 0.633
The examination of the No Opinions against the Nos lead to a p-value of 0.875, indicating that there is no significant difference between these two categories when taken as an indicator of gender, but when the No Opinions are compared to the Yes vote the p-value of 0.001 signifies an important difference in these two categories.
There is no statistical difference between No Opinion and No when it comes to predicting the gender of the respondent, but there is a difference when it comes to the Yes vote.
So while gender on its own is not a great leading indicator of the opinion that the subject will have, if the subject thinks that the contest should be abolished, then they are approximately 3 times more likely to be female, than if the person thinks that it should be allowed to continue (odds ratio=3.09), or if they had no opinion (odds ratio=2.996).
The examination of reason did not lead to the same clear outcome, due to technical problems with the statistical package (MINITAB©) used, when it comes to dealing with variables that have lots of zero values. However, by comparing the tabulated statistics, it is quite apparent that most who think that Miss World should be abolished do so on moral grounds (over 90% of Yes voters gave some moral reason as the main cause for their viewpoint)*, whereas a large percentage (42.42%) of those that think that it should not be abolished either don't care or feel that it should be allowed to continue because it doesn't affect them. This is followed by moral grounds - at 34.84% of the No votes. They think that it shouldn't be abolished because they believe in the freedom of speech, expression, and that the girls who participate should be allowed the right to choose to enter if they wish. 11% of male No voters gave some variation on the suimsuit contest as their main reason, but, not surprisingly, no females gave this as a reason. Less than 6% of respondents that gave No Opinion as their response to the question gave a reason other than they didn't know about it or that they just didn't care.
Overall the breakdown of the data was:
Male
Total 97 (55.11% of sample)
No 43 (44.33% of males, 24.43% of sample)
No opinion 37 (38.14% of males, 21.02% of sample)
Yes 17 (17.53% of males, 09.66% of sample)
Female
Total 79 (44.89% of sample)
No 23 (29.11% of females, 13.07% of sample)
No opinion 21 (26.58% of females, 11.93% of sample)
Yes 35 (44.30% of females, 19.89% of sample)
Totals
No 66 - 37.50%
No opinion 58 - 32.95%
Yes 52 - 29.55%
So we can see that the composition of the sample may have caused more problems if we had used methods such as linear regression, which would be more influenced by this thatn logistic regression techniques, when it comes to identifying group traits. Of course, it would have been nicer if the smple respondents had been less biased towards males aged 16-21. Among this category, the respondents were mostly students of the either science or engineering faculty, so, for this particular section of society, we have quite a reliable insight into how they feel on whether the Miss World contest should be abolished and why.
* For the purposes of the examination, and due to the somewhat obscure boundaries between them, moral, political and religious reasons were bundled into the same category.
Media section of the questionnaire
After the examination of the data, too many of the respondents gave the same answers to the prior knowledge of connected events in Nigeria last November to be able to come to any concrete results about how different forms of media created different viewpoints. Also, in the paper by Vincent Price and HJohn Zaller, Who gets the news? Alternative measures of news reception and their implications for research in PublicOpinion Quarterly: Volume 57, Number 2 (Summer 1993) pages 133-164, there are serious issurs raised when it comes to the reliability of self-reporting measures, when it comes to "discriminating who in fact gets the the news" as opposed to those who are simply exposed to the news, without necessarily learning and retaining the information. This is why the lack of variation in responses in the prior knowledge section, which aimed to distinguish between those who got the news and those who were merely exposed to the news, caused some problems when trying to determine any significant fluctutions within the data set.
Even when we assumed that we had a somewhat accurate representation of the knowledge level of the respondents, there was no statistically significant difference in the opinions in the various forms of media or of the various levels of knowledge about the November riots.
While disappointing, in that we have not gleamed much information, what did arise was that, given the media coverage of the event, and that most people claimed to watch/listen or read the news daily, there does seem to be some evidence, obtained by examining boxplots and the 95 percentiles for the median, that while the means of the knowledge level are similar for all three categories, the median value of the Yes respondents was higher - suggesting that, on the whole, those who said that the Miss World contest should be abolished were better informed about the events in Nigeria last November. On the whole, though, there was very little difference between the level of knowledge and the opinion formed.