Tools › Statistics › Surveys and Sampling

Surveys and Sampling

2 tools in the Statistics area.

Survey Analysis

Tabulation of survey responses pasted as respondent rows (a header line of column names, then one line per respondent; optional count and weight columns) or as a ready cross-tab of counts: for each chosen column a frequency table with weighted proportions, Wilson 95 % intervals and the margin of error; the overall margin of error at the design's sample size with the finite-population correction and the Kish design effect 1 + CV²(w) of the weights; Cochran's sample size for a target margin; a cross-tab of two columns with row and column percentages, Pearson χ², the likelihood-ratio G², Cramér's V, the adjusted standardized residuals and, for a 2 × 2 table, Fisher's exact test, the Yates-corrected χ² and the odds ratio; Likert items (mean, standard deviation, top-two-box, bottom-two-box, net score and the distribution) and the Net Promoter Score with its standard error and confidence interval. The default case is Agresti's gender × party identification table (1991 General Social Survey, χ² = 7.01, G² = 7.00 on 2 degrees of freedom) typed as counted rows, with an illustrative satisfaction item and a 0–10 recommendation score added to the rows.

4 worked examples reproduced, listed on the tool page.

Sample Size and Power

Sample size, power and detectable effect for the common study designs: one-sample, paired and two-sample means (exact noncentral t, Welch for unequal variances, z when the standard deviation is known), one and two proportions (normal approximation with and without the continuity correction, the arcsine form, the exact binomial test and Fisher's exact test by enumeration), a correlation (Fisher z), one-way ANOVA (noncentral F with Cohen's f), chi-square goodness of fit or independence (noncentral chi-square with Cohen's w), a regression slope, a one-sample variance (chi-square), survival by the log-rank test (Schoenfeld and Freedman events, accrual and follow-up to the number of patients), non-inferiority and equivalence (TOST, exact for means) for means and proportions, cluster designs (design effect from the intraclass correlation and cluster size) and survey sampling (margin of error with the finite-population correction, stratified allocation). Each case prints the formula with the substituted values, the distribution used, the sample size per group and in total, a table of sizes at other powers and levels, and power curves against the sample size and the effect size. The default case is a two-sample t test for a standardized difference of 0.5 at α = 0.05 and 80 % power: 64 per group.

1 worked example reproduced, listed on the tool page.

Other Statistics sections

Descriptive and Outliers · Hypothesis Tests · Regression and Modelling · Distributions · Forecasting · Uncertainty and Monte Carlo · Probability