marlio

AP Statistics, Unit 3

Exam words: Inference for Categorical Data: Proportions

The 46 words a written answer in this unit needs. For each one: what it means, a wording that loses the mark and why, and a sentence that uses it the way a full answer would.

Study these words as a deck

Flashcards, Learn, Test and Match in Cards, with your Still learning marks kept, and the booklet to print on A4 or US Letter. Sign in to open it.

The words

  1. sample proportion

    p^=xn\hat{p} = \dfrac{x}{n}

    The proportion of successes in a sample, p-hat, estimating the population proportion p.

    Loses the mark: “the proportion of the population”. The proportion in the whole population is p; p-hat is computed from the sample.

    Example: In a sample of 400 voters, 232 support the park, so the sample proportion is 232/400 = 0.58.

  2. population proportion

    pp

    The proportion of the whole population with a characteristic, written p; a fixed parameter.

    Loses the mark: “the proportion in the sample”. That is p-hat; the parameter is the proportion in the entire population.

    Example: We want to estimate the population proportion of all county voters who support the park.

  3. standard deviation of the sample proportion

    σp^=p(1−p)n\sigma_{\hat{p}} = \sqrt{\dfrac{p(1-p)}{n}}

    The typical distance between p-hat and p across samples: the square root of p(1 - p)/n.

    Loses the mark: “square root of np(1 - p)”. That is the standard deviation of the count of successes, not of the proportion.

    Example: With p = 0.55 and n = 400, the standard deviation of the sample proportion is the square root of (0.55)(0.45)/400, about 0.025.

  4. Large Counts condition

    np≥10np \ge 10 and n(1−p)≥10n(1-p) \ge 10

    np and n(1 - p) both at least 10, so p-hat is approximately Normal.

    Loses the mark: “n is at least 30”. The n of 30 guide is for means; proportions need at least 10 expected successes and 10 expected failures.

    Example: The Large Counts condition holds since 400(0.55) = 220 and 400(0.45) = 180 are both at least 10.

  5. approximately Normal

    Describes a distribution that a Normal model fits well enough to use for probabilities.

    Loses the mark: “the sample is normal”. Say which distribution is approximately Normal, such as the sampling distribution of p-hat, and which condition justifies it.

    Example: Because Large Counts holds, the sampling distribution of p-hat is approximately Normal with mean 0.55 and standard deviation 0.025.

  6. sampling distribution of a difference

    μp^1−p^2=p1−p2\mu_{\hat{p}_1 - \hat{p}_2} = p_1 - p_2, μxˉ1−xˉ2=μ1−μ2\mu_{\bar{x}_1 - \bar{x}_2} = \mu_1 - \mu_2

    The distribution of the difference between two independent statistics, such as p-hat1 - p-hat2.

    Loses the mark: “centered at p1 + p2”. The center of a difference is the difference of the parameters, p1 - p2.

    Example: The sampling distribution of a difference in mean commute times is centered at mu1 - mu2 = 5 minutes.

  7. standard deviation of a difference

    σxˉ1−xˉ2=σ12n1+σ22n2\sigma_{\bar{x}_1 - \bar{x}_2} = \sqrt{\dfrac{\sigma_1^2}{n_1} + \dfrac{\sigma_2^2}{n_2}}

    For independent samples, the square root of the sum of the two sampling variances.

    Loses the mark: “minus sign under the square root”. Variances add even for a difference, so every term under the root is added.

    Example: The standard deviation of a difference in mean commute times is the square root of (36/50 + 49/40), about 1.39 minutes.

  8. independent samples

    Two samples where individuals in one have no link to individuals in the other.

    Loses the mark: “independent because they are different sizes”. Independence depends on how the samples were chosen, not on their sizes.

    Example: The commute times came from independent samples of 50 drivers and 40 bus riders chosen separately.

  9. point estimate

    A single statistic used as the best guess for an unknown parameter.

    Loses the mark: “the point estimate is the parameter”. A point estimate comes from the sample and will usually miss the parameter by some amount.

    Example: The point estimate for the proportion of county voters who support the park is p-hat = 0.58.

  10. confidence interval

    An interval of plausible parameter values, point estimate plus or minus margin of error.

    Loses the mark: “95% of the data”. A confidence interval estimates a parameter; it does not contain a percent of the individual values.

    Example: We are 95% confident that the confidence interval from 0.532 to 0.628 captures the true proportion of county voters who support the park.

  11. confidence level

    The long-run percent of intervals, from repeated random samples, that capture the true parameter.

    Loses the mark: “95% probability that the true proportion is in the interval”. Once computed, an interval either captures the parameter or not; the 95% describes the method over many samples.

    Example: A 95% confidence level means that if we took many random samples and built an interval from each, about 95% of those intervals would capture the true proportion.

  12. margin of error

    ME=z∗×SEME = z^* \times SE

    The most a point estimate is likely to miss the parameter by through sampling variability.

    Loses the mark: “the margin of error accounts for bias”. It covers only random sampling variability; a biased method stays biased however small the margin.

    Example: The margin of error of 0.048 means the poll's estimate is likely within about 5 percentage points of the true proportion, ignoring bias.

  13. standard error

    SEp^=p^(1−p^)nSE_{\hat{p}} = \sqrt{\dfrac{\hat{p}(1-\hat{p})}{n}}

    An estimate, computed from sample data, of a statistic's standard deviation.

    Loses the mark: “the standard deviation of the data”. The standard error measures how much the statistic varies across samples, not how individual values vary.

    Example: The standard error of p-hat is the square root of (0.58)(0.42)/400, about 0.0247.

  14. critical value

    z∗z^*

    The multiplier from a Normal or t distribution giving the margin's size in standard errors.

    Loses the mark: “always use 1.96”. 1.96 is only for 95% confidence; use 1.645 for 90% and 2.576 for 99%.

    Example: For 99% confidence the critical value is z* = 2.576, so the interval is wider than the 95% interval.

  15. sample size

    nn

    The number of individuals in a sample, n; larger n gives a smaller standard error.

    Loses the mark: “a larger sample size removes bias”. More data shrinks random variability only; a biased method stays biased however big the sample.

    Example: Quadrupling the sample size cuts the margin of error in half at the same confidence level.

  16. one-sample z interval for a proportion

    p^±z∗p^(1−p^)n\hat{p} \pm z^*\sqrt{\dfrac{\hat{p}(1-\hat{p})}{n}}

    A confidence interval for p: p-hat plus or minus z* times the standard error.

    Loses the mark: “t interval for a proportion”. Proportions use z*; t is for means when sigma is unknown.

    Example: A one-sample z interval for a proportion gives 0.58 plus or minus 1.96(0.0247), or 0.532 to 0.628.

  17. Random condition

    A check that data come from random sampling or random assignment, so inference is valid.

    Loses the mark: “the sample is large so it is random”. Size does not make a sample random; name the random sampling or assignment the study used.

    Example: The Random condition is met because the 400 voters were a simple random sample from the county voter list.

  18. plausible value

    A value of the parameter inside the confidence interval, so consistent with the data.

    Loses the mark: “the interval proves”. An interval gives plausible values; it does not prove the parameter is any one of them.

    Example: Since 0.50 is below the interval 0.532 to 0.628, 0.50 is not a plausible value for the true proportion of supporters.

  19. significance test

    A procedure using sample data to weigh evidence against a null hypothesis about a parameter.

    Loses the mark: “the test proves the claim”. A test measures evidence; it never proves a hypothesis true or false.

    Example: A significance test tells us whether the sample gives convincing evidence that more than half of voters support the park.

  20. null hypothesis

    H0H_0

    The claim of no effect or no difference that a test weighs evidence against.

    Loses the mark: “H0: p-hat”. Hypotheses are about parameters such as p or mu, never about sample statistics.

    Example: The null hypothesis is H0: p = 0.50, where p is the true proportion of county voters who support the park.

  21. alternative hypothesis

    HaH_a

    The claim a test seeks evidence for, that the parameter differs from the null value.

    Loses the mark: “Ha: x-bar”. State the alternative about the parameter, not about the sample statistic.

    Example: The alternative hypothesis is Ha: p > 0.50, since the group claims majority support.

  22. p-value

    Assuming H0 is true, the probability of a statistic at least as extreme as observed.

    Loses the mark: “probability that the null hypothesis is true”. The p-value is computed assuming H0 is true; it is not the probability that H0 is true.

    Example: Assuming the true proportion is 0.50, the p-value of 0.0007 is the probability of a sample proportion of 0.58 or higher in a random sample of 400.

  23. significance level

    α\alpha

    The cutoff alpha: p-values below it reject H0; also the Type I error probability.

    Loses the mark: “the p-value is the significance level”. Alpha is chosen in advance; the p-value is computed from the data and compared with it.

    Example: At the significance level alpha = 0.05, a p-value of 0.0007 leads us to reject H0.

  24. statistically significant

    Describes a result with a p-value below the significance level, so unlikely under H0.

    Loses the mark: “significant means important”. Statistical significance is about evidence against H0, not about the size or importance of an effect.

    Example: The difference in recovery rates is statistically significant at alpha = 0.05, even though it is only 2 percentage points.

  25. fail to reject the null hypothesis

    The decision when the p-value is not below alpha: no convincing evidence for Ha.

    Loses the mark: “accept the null hypothesis”. Not finding evidence against H0 does not show H0 is true; say we fail to reject H0.

    Example: With a p-value of 0.21, above 0.05, we fail to reject the null hypothesis: there is not convincing evidence that the coin favors heads.

  26. reject the null hypothesis

    The decision when the p-value is below alpha: convincing evidence against H0 and for Ha.

    Loses the mark: “proves the alternative”. Rejecting H0 gives strong evidence for Ha, not proof; sampling variability could still mislead.

    Example: Because the p-value of 0.0007 is less than 0.05, we reject the null hypothesis.

  27. convincing evidence

    Conclusion wording: the data do or do not give convincing evidence for the alternative hypothesis.

    Loses the mark: “evidence for the null hypothesis”. A test only weighs evidence against H0; a large p-value is a lack of evidence, not evidence for H0.

    Example: There is convincing evidence that the true proportion of county voters who support the park is greater than 0.50.

  28. Type I error

    Rejecting H0 when H0 is actually true; its probability equals the significance level alpha.

    Loses the mark: “Type I error is failing to reject a false”. That describes a Type II error; a Type I error rejects a true H0.

    Example: A Type I error here would be concluding that most voters support the park when in fact the true proportion is 0.50.

  29. Type II error

    β\beta

    Failing to reject H0 when the alternative is actually true; its probability is called beta.

    Loses the mark: “Type II error is rejecting a true”. That is a Type I error; a Type II error misses a real effect by failing to reject a false H0.

    Example: A Type II error would be failing to find convincing evidence of majority support when most voters really do support the park.

  30. power

    1−β1 - \beta

    The probability a test correctly rejects H0 when a specific alternative is true, 1 - beta.

    Loses the mark: “power is the chance of a Type I error”. That is alpha; power is the chance of rejecting H0 when H0 is false.

    Example: Doubling the sample to 800 voters raises the power of the test to detect a true proportion of 0.55.

  31. one-sample z test for a proportion

    z=p^−p0p0(1−p0)/nz = \dfrac{\hat{p} - p_0}{\sqrt{p_0(1-p_0)/n}}

    A test of H0: p = p0 using z and a p-value from the standard Normal.

    Loses the mark: “standard error uses p-hat in the test”. In a test the standard error uses the null value p0, since we assume H0 is true.

    Example: A one-sample z test for a proportion gives z = (0.58 - 0.50)/0.025 = 3.2 for the park poll.

  32. test statistic

    A standardized value giving how many standard errors the statistic lies from the null value.

    Loses the mark: “the test statistic is the p-value”. The test statistic measures distance from the null value; the p-value is the tail area beyond it.

    Example: The test statistic z = 3.2 says the sample proportion is 3.2 standard errors above the null value of 0.50.

  33. null value

    p0p_0

    The parameter value stated in H0, such as p0, assumed true to find the p-value.

    Loses the mark: “Large Counts with n times p-hat”. For a test, check Large Counts with the null value: n times p0 and n(1 - p0) both at least 10.

    Example: With the null value p0 = 0.50, the expected counts are 400(0.50) = 200 successes and 200 failures.

  34. two-sided test

    A test where the alternative is not equal, so its p-value uses both tails.

    Loses the mark: “one tail for not equal”. For a not-equal alternative, count both tails: double the one-tail area.

    Example: A two-sided test of H0: p = 0.25 against Ha: p is not 0.25 gives a p-value of 2(0.04) = 0.08.

  35. one-sided test

    A test whose alternative says the parameter is greater, or less, than the null value.

    Loses the mark: “double the p-value for a one-sided test”. Doubling is only for two-sided alternatives.

    Example: Since the claim is that more than half support the park, we use a one-sided test with Ha: p > 0.50.

  36. two-sample z interval for a difference in proportions

    (p^1−p^2)±z∗p^1(1−p^1)n1+p^2(1−p^2)n2(\hat{p}_1 - \hat{p}_2) \pm z^*\sqrt{\dfrac{\hat{p}_1(1-\hat{p}_1)}{n_1} + \dfrac{\hat{p}_2(1-\hat{p}_2)}{n_2}}

    A confidence interval for p1 - p2 using z* and the unpooled standard error.

    Loses the mark: “pooled proportion in the interval”. An interval does not assume p1 = p2, so it uses each sample's own p-hat.

    Example: A 95% two-proportion z interval for p1 - p2 is 0.03 to 0.15, so every plausible difference favors the new reminder text.

  37. two-sample z test for a difference in proportions

    A test of H0: p1 = p2 using z and the pooled proportion.

    Loses the mark: “H0: p-hat1 = p-hat2”. State hypotheses about the population proportions p1 and p2, not the sample proportions.

    Example: A two-proportion z test of H0: p1 = p2 gives z = 2.9 and a two-sided p-value of 0.004 for the reminder experiment.

  38. pooled proportion

    p^c=x1+x2n1+n2\hat{p}_c = \dfrac{x_1 + x_2}{n_1 + n_2}

    The combined proportion of successes in both samples, used when testing H0: p1 = p2.

    Loses the mark: “average of the two sample proportions”. Pool the counts, total successes over total sample size, which differs from a plain average unless n1 = n2.

    Example: With 90 of 300 and 60 of 300 responding, the pooled proportion is 150/600 = 0.25.

  39. chi-square test for goodness of fit

    A test of whether one categorical variable's population distribution matches a claimed set of proportions.

    Loses the mark: “the distribution is correct”. A large p-value only means no convincing evidence against the claimed proportions, not that they are right.

    Example: A chi-square test for goodness of fit checks whether the four candy colors occur in the claimed proportions 0.40, 0.30, 0.20 and 0.10.

  40. expected count

    The count a category or cell would have on average if H0 were true.

    Loses the mark: “observed counts are at least 5”. The condition is on the expected counts, which must all be at least 5.

    Example: With 200 candies, the expected count for red is 200(0.40) = 80.

  41. observed count

    The actual number of individuals falling in a category or cell of the sample data.

    Loses the mark: “observed percentages”. The chi-square statistic is built from counts, not percents or proportions.

    Example: The observed count for red was 92, compared with an expected 80.

  42. chi-square statistic

    χ2=∑(O−E)2E\chi^2 = \sum \dfrac{(O - E)^2}{E}

    The sum over all categories or cells of (observed - expected) squared divided by expected.

    Loses the mark: “divide by the observed count”. Each term divides by the expected count.

    Example: The chi-square statistic for the candy colors is 6.4, the sum of the four (O - E)^2/E components.

  43. chi-square distribution

    df=k−1df = k - 1 (goodness of fit)

    A right-skewed distribution of nonnegative values, set by degrees of freedom, used for chi-square p-values.

    Loses the mark: “two-tailed chi-square”. Only large chi-square values count against H0, so the p-value is the right-tail area.

    Example: With 4 colors, the chi-square distribution with 3 degrees of freedom gives P(chi-square > 6.4) = 0.094.

  44. chi-square test for independence

    E=(row total)(column total)grand totalE = \dfrac{(\text{row total})(\text{column total})}{\text{grand total}}, df=(r−1)(c−1)df = (r-1)(c-1)

    A test of whether two categorical variables are associated in one population, using one sample.

    Loses the mark: “proves the variables are independent”. A large p-value is only a lack of evidence of association, not proof of independence.

    Example: A chi-square test for independence on a random sample of 500 adults asks whether preferred news source and age group are associated.

  45. chi-square test for homogeneity

    A test of whether a categorical variable has the same distribution in several independent groups.

    Loses the mark: “homogeneity uses different degrees of freedom”. Homogeneity and independence tests use identical arithmetic, with df = (r - 1)(c - 1).

    Example: A chi-square test for homogeneity compares the distribution of favorite cafeteria meals among separate random samples of 9th, 10th and 11th graders.

  46. chi-square component

    One cell's contribution, (O - E)^2/E, to the chi-square statistic; larger ones show where data differ.

    Loses the mark: “the largest count shows the difference”. Large counts are not the point; compare each cell's (O - E)^2/E to see where the mismatch is.

    Example: The largest chi-square component, 5.2, comes from older adults who prefer print news, far more of them than expected.