Confidence Intervals Around a proportion Significance Tests Not Every Difference Counts Difference in Proportions Difference in Means
The 95% confidence interval around a sample proportion is: And the 99.7% confidence interval would be:
The margin of error is calculated by:
In a poll of 505 likely voters, the Field Poll found 55% support for a constitutional convention.
The margin of error for this poll was plus or minus 4.4 percentage points. This means that if we took many samples using the Field Poll’s methods, 95% of the samples would yield a statistic within plus or minus 4.4 percentage points of the true population parameter.
Perhaps the differences in our dependent variable that we observe from group to group in our sample could be due to random chance alone. A “significance test” asks: How likely is it that we would observe a difference that large in our sample, if there were no difference in the population?
Significance tests begin with the “null hypothesis,” the claim that there is no difference in the dependent variable from one group to another (measured as a mean or a proportion). This is akin to saying that the independent variable that we use to group cases has no effect on the dependent variable.
Mathematically, the null hypothesis says: And the sample difference in means is normally distributed with a mean equal to the population difference in means with a standard error equal to:
For interval and ratio dependent variables, you can conduct a difference in means test: Where
For nominal and ordinal dependent variables, you can conduct a difference in proportions test: Where If the confidence interval does NOT include zero, we can reject the null hypothesis with 95% confidence.
60.4% of the 1024 white residents in my survey voted, as did 24.5% of the 625 Latinos. This test tells us we can reject the null hypothesis.