Summary
Highlights
Unit 1 covers the distinction between categorical and quantitative data. Categorical data, like eye color, is best represented using two-way tables, which illustrate intersections between variables. Key terms include marginal, joint, and conditional relative frequencies. Quantitative data, such as heights, is described using the C-SOCKS acronym (Context, Shape, Outliers, Center, Spread), focusing on descriptive language. Important basic terms include mean, standard deviation (and how to interpret it in context), median, and range. Box plots, a visual representation, are constructed using a five-number summary: minimum, Q1 (25th percentile), median (50th percentile), Q3 (75th percentile), and maximum. The IQR (Q3 - Q1) is crucial for identifying outliers using specific formulas (Q1 - 1.5 * IQR for low-end, Q3 + 1.5 * IQR for high-end). Other concepts include percentiles (percentage of values less than or equal to a specific value), cumulative relative frequency, and z-scores (number of standard deviations from the mean). Data transformations (adding/subtracting or multiplying/dividing constants) affect the center and/or variability but preserve the shape. The unit concludes with density curves and normal distributions, emphasizing the 68-95-99.7 rule and calculator commands (normPDF, normCDF, inverse normal). Normal probability plots are also introduced to assess how well data fits a normal distribution.
Unit 2 focuses on describing scatter plots and regression analysis. Scatter plots are described using the acronym 'Seed' (Context, Direction, Outliers, Form, Strength). The correlation coefficient (R-value) measures the strength and direction of linear relationships, ranging from -1 to 1. Changes in units or switching axes do not affect the R-value. Outliers can strengthen or weaken 'R' depending on their position relative to the data pattern. A critical takeaway is that correlation does not imply causation. Regression lines, or 'best-fit' lines (ŷ = a + bx), are used to estimate values. Residuals represent the error in prediction (actual y - predicted y), with negative residuals indicating overestimation and positive indicating underestimation. Interpreting the slope and y-intercept of the regression line in context is vital. The least squares regression line minimizes the sum of squared residuals. Key metrics from computer printouts include the S-value (average distance of predicted values from the line) and R-squared (coefficient of determination, percentage of response variable variance explained by the explanatory variable). Outliers can influence the least squares regression line, affecting the slope and y-intercept differently based on their position. Residual plots are used to assess the appropriateness of a linear model: a clear pattern suggests a linear function is not the best fit, while an unclear, random pattern suggests it is.
Unit 3 delves into various sampling methods and experimental designs. Common sampling methods include Simple Random Sample (SRS), Stratified Random Sample (splitting population into homogeneous groups and sampling from each), Cluster Sample (splitting into heterogeneous groups and sampling entire clusters), and Systematic Random Sample (sampling at regular intervals after a random start). Bad sampling methods include Convenience Sample (choosing easily accessible individuals) and Voluntary Response Sampling (allowing self-selection, leading to bias). Shortcomings in sampling include undercoverage (some groups excluded), non-response (selected individuals don't respond), response bias (false/misleading answers), and wording bias (poorly phrased questions influencing answers). The unit distinguishes between observational studies (observing without influencing) and experiments (manipulating variables to observe effects). Experiments should adhere to four principles: Comparison, Random Assignment, Control, and Replication. Key vocabulary includes factor (explanatory variable), level (specific value of a factor), confounding (other variables affecting results), placebo (fake treatment), single-blind (subjects unaware of treatment), and double-blind (neither subjects nor researchers aware). Experimental designs covered are Randomized Block Design (grouping by similar characteristics and randomizing treatments within blocks) and Matched Pairs Design (pairing subjects and randomly assigning treatments or controls within pairs).
Unit 4 explores probability and random variables. Probability is the chance of an event, expressed as 0 to 1, unpredictable in the short term but predictable in the long term through many trials. Simulations model real-world events to estimate probabilities using a four-step process. Key probability rules involve mutually exclusive events (no overlap) and independent events (outcome of one doesn't affect another). Formulas for 'either/or' (addition rule) and 'both' (multiplication rule) events are discussed, along with the complement rule (probability of an event not occurring). Visualizing probability can be done using Venn diagrams, two-way tables, and probability trees. Random variables are categorized into discrete (specific, countable values), continuous (any value within a range), binomial (fixed trials, success/failure, independent, set probability), and geometric (number of trials until first success). The mean and standard deviation for discrete/continuous variables can be found using calculator functions; calculating the expected value for a discrete variable is also demonstrated. Transformations of probability distributions (adding/subtracting or multiplying/dividing) affect the center and/or variability while preserving the shape. Formulas for means and standard deviations when adding or subtracting independent random variables are presented. The 'BINS' acronym (Binary, Independent, Number of trials fixed, Success probability same) is used to identify binomial settings, with corresponding calculator commands (binomPDF, binomCDF). Geometric random variables are identified by focusing on the 'first success' with calculator commands (geometPDF, geometCDF) and formulas for mean and standard deviation. The critical distinction between binomial and geometric is the fixed number of trials in binomial vs. 'until' the first success in geometric.
Unit 5 covers sampling distributions, a foundational concept for the rest of AP Statistics. It differentiates between a statistic (number from a sample) and a parameter (number from a population). A sampling distribution is a probability distribution of a statistic obtained through repeated sampling. An unbiased estimator's sampling distribution mean equals the population parameter. The unit details sampling distributions for proportions: taking repeated samples, finding each sample's proportion, and plotting them. Key formulas for the mean and standard deviation of sampling distributions for proportions are provided, noting that increasing sample size decreases variability. Crucial assumptions and conditions for inference are introduced: random sampling, the 10% condition (sample size is less than 10% of population to assume independence), and the large counts condition (at least 10 successes and 10 failures). If these conditions are met, the sampling distribution is approximately normal, allowing for Z-value calculations and probability determination using normal CDF or Table A. Similarly, for means, repeated samples are taken, their means are plotted, and formulas for the mean and standard deviation of sampling distributions for means are given. The same conditions (random sampling, 10% condition, large enough sample size) apply. The Central Limit Theorem (CLT) states that if the sample size is at least 30, the sampling distribution of the mean will be approximately normal, even if the population distribution is not. Z-values and probabilities are calculated similarly to proportions.
Unit 6 focuses on inference for categorical data, specifically proportions. A point estimate is a statistic used to estimate a population parameter. The unit distinguishes between a confidence interval (range of values for the parameter) and a confidence level (probability the parameter falls within the range). Proper interpretation of both is crucial, emphasizing context. Key caveats include: increasing confidence level widens the interval, increasing sample size narrows it, and bias does not affect the margin of error. The 'PANIC' acronym (Parameter, Assumptions, Name of test, Interval, Conclude) guides confidence interval construction. For proportions (one or two samples), the parameter is the true proportion, the procedure is a Z-interval, and conditions include random sampling, 10% independence, and large counts for both samples if applicable. Interval calculation can use a calculator or the formula: point estimate ± margin of error (Z* × standard deviation). The conclusion must be in context. A key note for two-sample confidence intervals: if zero is included, a difference in population proportion cannot be concluded. For significance tests, the 'PHANTOMS' acronym (Parameters, Hypothesis, Assumptions, Name of test, Test statistic, Obtain p-value, Make decision, State conclusion) is used. Hypotheses (null and alternative) are defined. Conditions are similar to confidence intervals. The test statistic and p-value are typically obtained via calculator. The p-value dictates the decision to reject or fail to reject the null hypothesis, often comparing it to an alpha level (e.g., 0.05). Conclusions are stated in context, referring to the alternative hypothesis. The unit also introduces Type I error (rejecting a true null hypothesis, probability is alpha) and Type II error (failing to reject a false null hypothesis). Power is the probability of correctly rejecting a false null hypothesis (1 - Type II error probability). Power can be increased by increasing alpha, increasing sample size, or increasing the distance between the null and alternative hypothesis values.
Unit 7 extends inference to quantitative data, specifically means. Building on Unit 6, the primary change is using a critical t-value (t*) instead of a z-value. The concept of degrees of freedom (df = n - 1) is introduced for calculating t*. The PANIC acronym guides confidence interval construction: parameter of interest is the true mean, the procedure is a one-sample or two-sample t-interval, and assumptions/conditions are similar to proportions but with a modification for normality. Random sampling and the 10% condition remain. For normality, one of three conditions must be met: the population is already normal, the sample size is >= 30 (Central Limit Theorem), or the sample data shows no strong skewness or outliers. These conditions apply to both samples in a two-sample test. The interpretation of the confidence interval is for the true mean or the true difference in means, in context. Significance tests for means also follow the PHANTOMS acronym. Hypotheses relate to the true mean. The procedure is a one-sample or two-sample t-test. Assumptions and conditions are identical to those for t-intervals. The conclusion is based on comparing the p-value to the alpha level, making a decision to reject or fail to reject the null hypothesis, and stating the conclusion in context, specifically addressing the alternative hypothesis. The phrasing for rejecting/failing to reject and concluding/not concluding the alternative hypothesis is emphasized for proper interpretation.
Unit 8 covers inference for categorical data using chi-square tests, a smaller but important part of the AP exam, often appearing in multiple-choice questions. Three types are discussed: Chi-Square Test for Goodness-of-Fit, Chi-Square Test for Homogeneity, and Chi-Square Test for Association/Independence. For Goodness-of-Fit, the key question is whether an observed distribution significantly differs from an expected distribution (e.g., M&M color distribution). This test does not use a parameter of interest or sample statistic. It follows PHANTOMS, with hypotheses stating the expected distribution. Conditions include random sample, 10% condition, and large counts (all expected counts >= 5). Degrees of freedom equal the number of categories minus one. The test statistic indicates discrepancy from expected values; a higher statistic means greater difference. The p-value determines whether to reject the null hypothesis. The Chi-Square Test for Homogeneity assesses if two or more populations share the same distribution for a single categorical variable. It's identified by having multiple samples and one variable. The null hypothesis states no difference in variables between groups. Degrees of freedom and expected counts are calculated based on rows and columns of a two-way table. The Chi-Square Test for Association/Independence determines if an association exists between two categorical variables within a single sample. The null hypothesis states no association. All other aspects (conditions, test statistic, p-value, conclusion) are similar across all chi-square tests, with context-specific adjustments in the hypotheses and conclusions. Notably, degrees of freedom for homogeneity and association tests are (rows - 1) * (columns - 1).
Unit 9, the final unit of AP Statistics, focuses on inference for slope, similar to previous inference units but applied to the slope (beta) of the least squares regression line. The goal is to determine if a linear relationship exists between two variables. Data collection involves a random sample, identifying explanatory (independent) and response (dependent) variables (e.g., study time and test scores). A confidence interval for slope follows the PANIC acronym: the parameter is the true slope (beta) of the population least squares regression line. There are five crucial assumptions/conditions (usually why it's not a full FRQ task): 1. Linear relationship (checked via scatter plot or residual plot), 2. Independence (10% condition for sampling, or independent observations for experiments), 3. Normal distribution of residuals (no strong skewness or outliers in residual plot/dot plot), 4. Equal standard deviation of residuals (residual plot shows no increasing/decreasing variance), and 5. Random sampling/assignment. The test name is a t-interval for the slope. The interval is calculated as the estimated sample slope (b) ± (t* × standard error of the slope), with degrees of freedom n-2. The conclusion states confidence that the interval captures the true slope in context. A significance test for slope uses PHANTOMS. Hypotheses typically set the null hypothesis (Ho) that the slope (beta) equals zero (no linear relationship), while the alternative hypothesis (Ha) can be greater than, less than, or not equal to zero. The test name is a t-test for the population slope. The test statistic is (estimated sample slope - hypothesized population slope) / standard error of the slope. Obtaining the p-value requires raw data for calculator use (t-test for slope) or manual calculation using the t-value, degrees of freedom (n-2), and tcdf. The conclusion is based on comparing the p-value to the alpha level, deciding to reject or fail to reject Ho, and concluding (or not concluding) that a linear relationship exists between the two variables, all in context.