Sunday, March 16, 2014

AP Statistics Ch 12: Significant Tests in Practice

Part 1: Population Mean

  • use a t-score for test statistic
    • t = (x-bar - mean value claimed) / (standard deviation from data / sqrt(sample size))
    • simplifies to t = (x-bar - mean value claimed) / (s/sqrt(n))
    • measures how far x-bar is from the claimed mean; measured in standard deviation
  • types of alternative hypothesis
    • meana > mean0
    • meana < mean0
    • meana does not equal to mean0 (this is two-sided)
  • still need to have Normality, SRS, and Independence to use it
  • for paired t-tests, carry out the same steps, only this time, the parameter is the difference between the two samples
  • Skewed curve is the most serious of non-Normality
    • larger sample size (over 30)>>>> Central Limit Theorem; can count as Normal
  • high power is important; detects deviation from the claimed parameter value
    • is probability that test will reject the null hypothesis when the mean is a certain alternative value
    • 1 - Probability of Type 2 error

Part 2: Population Proportion

  • sample must be Normal, Independent, and from SRS
  • Will use z-test
    • z = (p-hat - p0)/ (sqrt((p0(1 - p0))/n)
      • p-hat is the p from the alternative hypothesis
      • p0 is the p from the null hypothesis
      • n is the sample size
  • types of alternative hypothesis
    • pa > p0
    • pa < p0
    • pa does not equal to p0 (this is two-sided)
  • aka large-sample test
    • based on Normal curve; p gets more accurate as the sample size increases
  • confidence intervals tell us if our test is good enough to detect deviations; if confidence interval really large, then the power of the test is small >>> hard to detect deviations, consider increasing sample size and other actions that make the power larger
    • also tells us how far above or below the value reported in the null hypothesis the real value really is; good to include in conclusion
    • standard error for confidence interval is
      • sqrt((pa(1 - pa))/n
        • pa is value reported in sample (which is the data)
        • n is the sample size
    • doing this will mean that significance level of two-tailed significance test is not exactly the 1 - significance value of confidence interval of two-tailed significance test, but it is close enough

No comments:

Post a Comment