SAS programming Statistical Analyst Certification Course

Access SAS software with SAS on demand for academics, a cloud-based SAS studio; create a profile, verify your email, sign in, and upload course material and files to home folder.

upload data files and sas programs into sas on demand for academics by extracting zip files and organizing them in a dedicated sas base codes folder.

Learn to update SAS programs for SAS on demand for academics by correcting home directory paths, saving updated files as new source programs, and validating runs.

Learn to perform two sample t tests for independent groups using pooled variance when variances are equal, verify assumptions with F tests and QQ plots in SAS proc t test.

Explore how ANOVA identifies relationships between a categorical predictor and a numeric outcome, and generate predicted values and residuals with the output statement in proc glm.

Explore the normality assumption for Anova by using the plots option in proc glm to generate diagnostic plots, including the normal quantile plot and histogram of residuals.

Use Levene’s test to check homogeneity of variances in anova and apply the means statement in proc glm; the example shows p=0.09, not rejected.

Explore the randomized block design to control nuisance factors and improve treatment effect estimates. Apply SAS GLM to test reading score differences across parent education groups, blocking by SMS minutes.

This tutorial demonstrates post hoc analysis for a randomized block design using the lsmeans statement with Tukey adjustment, comparing group means and identifying significant differences.

Assess normality of errors, equal block variance, and no interaction in a randomized block design using plots in proc glm, with normal quantile plots, histograms, and kolmogorov-smirnov tests on residuals.

In unbalanced randomized block designs, use type iii sums of squares and the lsmeans statement for one-factor analysis, and rely on unweighted means for accurate pairwise comparisons via proc glm.

Explore two-factor analysis of variance, a two-way ANOVA with interaction, covering main effects, the interaction concept, and how to model factor A, factor B, and their interaction.

Analyze a two-factor ANOVA on reading scores by parent education level and state cluster. Find a significant interaction and improved model fit, with interpretable least squares means across six groups.

Learn how to prepare input variables for predictive modeling by cleaning data, handling missing values, reducing redundancy, and ensuring preprocessing is applied to training data to prevent information leakage.

Learn to handle missing values with complete case analysis and imputation, using missing value indicators and median imputation, while considering MCAR, MAR, and NMAR patterns in SAS.

Use the Greenacre method to collapse state categories, import Excel data with proc import, compute chi-square via proc freq, and use proc means to create state proportions for clustering.

Determine the optimal cluster count by evaluating chi-square p values and their log p values across seven steps, selecting the cutoff where the reduction is minimal, here two clusters.

Identify a two-cluster solution using proc sql and proc tree, then create a sta_class variable to compare chi square via the Greenacre method.

Apply variable clustering to reduce large input sets by addressing redundancy and relevancy using the VARCLUS procedure. Group highly correlated variables into clusters and select representatives via rotated principal components.

Use sas proc varclus to reduce input redundancy by clustering correlated variables. Select twitter minute as the cluster representative and create gender and state dummies, reporting variance explained.

Screen variables with the score and class procedures, assess relevancy using Spearman and Hoeffding's d, detect non-linear associations, and compare with Pearson via rank scatter plots.

Explore sas proc corr to detect associations using spearman and hoeffding correlations on a reduced input set and a target variable, with data prep and ods outputs.

Transform and merge Spearman and Hoeffding one-row outputs into a final data set, creating eight variable entries with correlation coefficients, p values, and ranks.

Screen inputs for monotonic relevance using Spearman and Hoeffding D, flagging variables with p values over 0.5 as potentially irrelevant and noting any non-monotonic patterns for further empirical logit analysis.

Apply empirical logit to detect non-linear associations in logistic regression, using logit plots by predictor and binning via the rank procedure and proc means to reveal nonlinearity.

Explore the relationship between two continuous variables with scatter plots and correlation coefficients, and learn how exploratory data analysis, regression lines, influential data points, and outliers affect interpretation.

Quantify the association between two continuous variables with the Pearson correlation using the CORR procedure, interpreting strength, direction, and significance alongside scatter plots and descriptive statistics.

Compare measures of fit in multiple linear regression using R-squared and adjusted R-squared, noting that adding predictors inflates R-squared, four predictors can match seven in adjusted R-squared and rmse.

Quantify a predictor's unique impact in a linear regression using squared semi partial correlation to reveal the share of R-squared explained by each predictor for read score beyond others.

This lecture covers detecting collinearity in regression using variance inflation factor and the condition index, with collin and collinoint, and advises deleting predictors or applying ridge or principal component regression.

Learn backward elimination in proc reg to remove the least significant predictors by p-values, ending when remaining variables meet p<0.01 and observing adjusted r-squared trends.

Explore forward selection in proc reg by selecting predictors by p-value and r-squared, stopping at 0.01, with final model including rate time minute update, Twitter minute, and status cluster.

Explore the glmselect procedure for variable selection, highlighting stepwise methods, collinearity cautions, and p-value risks; review a SAS glmselect example predicting reading scores.

Explore the four linear regression assumptions—linearity, normality of errors, and constant, independent variance—and use residual plots and transformations or generalized linear models to address violations.

Explore how to check multiple linear regression assumptions in SAS using PROC REG, generating residual plots, diagnostics, and normality checks to verify variance, independence, and model fit.

Explore logistic regression with a continuous predictor, interpret odds ratios, and assess model fit with AIC, likelihood ratio tests, and confidence intervals; evaluate predictive power using ROC AUC.

Learn to generate the data set and create odds ratio plots with 95% confidence intervals using proc logistic, including plots for unit one and unit ten.

Explore logistic regression with a three-level categorical predictor using effect coding parameterization in SAS, including dummy design variables, class statement handling, and interpretation via odds ratios.

Learn to run multiple logistic regression in SAS with proc logistic, using a pedu class variable, report odds ratios, and identify predictors such as read time and Twitter.

Master backward elimination in multiple logistic regression with SAS PROC LOGISTIC, starting from a full model and using chi-square tests to remove least significant predictors at a 0.01 level.

Learn how to customize variable selection in multiple logistic regression with forward, backward, and stepwise methods using start, stop, and include options to shape the final model.

Explore best subset selection for multiple logistic regression, using score-based chi-square criteria to choose models from two to six predictors, including dummy variables from a categorical predictor.

Model interaction terms in multiple logistic regression and test their significance. Apply backward elimination under model hierarchy and interpret conditional odds ratios with plots.

Score new data with logistic regression using outmodel and inmodel options (and opmode) in proc logistic, producing p0 and p1 predictions, and note rare-event oversampling requires intercept adjustment before scoring.

Use proc surveyselect to create stratified training and validation data sets by 70/30 splits on the target variable, then verify assignments with a selected indicator.

Evaluate classifier performance using the classification table, ROC curve, and gains and lift charts; compare models with hit rate, sensitivity, precision, and specificity.

Assess classifier performance against the naive rule and ensure generalizability by balancing training and validation hit rates to avoid overfitting.

Compare seven-predictor and three-predictor models using the roc and roc contrast statements on validation data set, evaluating roc curves and AUC, and prefer the simpler model when differences are non-significant.

Explains how lift charts measure classifier performance for targeting observations, showing lift as the ratio to random performance across depth or cutoff, with SAS plotting steps and reference lines.

Learn to score validation data using adjusted posterior probabilities to account for oversampling of rare events in SAS logistic regression, and compare manual adjustment, offset intercept, and p event methods.

Apply a manually adjusted intercept via an offset to correct biased population priors in oversampled data, enabling unbiased posterior probabilities in logistic regression for validation data.