From August 15^th to 19^th, our fan­tas­tic quan­ti­ta­tive method­ol­o­gists will offer five full-day hybrid (in-per­son and online) cours­es on advanced top­ics in sta­tis­ti­cal mod­el­ing, mea­sure­ment, and infer­ence. All ses­sions will be based in the freely avail­able R sta­tis­ti­cal soft­ware environment.

The training has five courses:

Mon­day, 8/15, 9:00 am – 4:00 pm, Townsend 222

Scale Devel­op­ment (Dr. Wes Bonifay)

In this ses­sion, Dr. Boni­fay will pro­vide an overview of the entire psy­cho­me­t­ric test (survey/ scale/ ques­tion­naire) con­struc­tion process. This course will empha­size mod­ern valid­i­ty the­o­ry – as rec­om­mend­ed by the APA/AERA/NCME Stan­dards for Edu­ca­tion­al & Psy­cho­log­i­cal Test­ing – in every phase of this process. Atten­dees will learn 1) to design a test blue­print that will guide and facil­i­tate item writ­ing, 2) to incor­po­rate valu­able feed­back from sub­stan­tive experts and mem­bers of the tar­get pop­u­la­tion of test-tak­ers, 3) to use cut­ting-edge sta­tis­ti­cal meth­ods (e.g., item response the­o­ry) to ana­lyze and refine the ini­tial test data, 4) to estab­lish test fair­ness and inves­ti­gate the con­se­quences of test use, and 5) to eval­u­ate, final­ize, dis­sem­i­nate, and main­tain the test. No pre­vi­ous test con­struc­tion expe­ri­ence is necessary.

Tues­day, 8/16, 9:00 am – 4:00 pm, Townsend 222

Prac­ti­cal Mul­ti­level Mod­el­ing (Dr. Fran­cis Huang) 

Mul­ti­level mod­el­ing (MLM) as an ana­lyt­ic tech­nique for the analy­sis of clus­tered data (e.g., stu­dents with­in schools, patients with­in hos­pi­tals) has grown over the years. The work­shop will intro­duce applied researchers to basic MLM con­cepts using R.  Atten­dees will learn: 1) how to con­struct var­i­ous types of mul­ti­level mod­els (i.e., uncon­di­tion­al, ran­dom inter­cept, ran­dom slope mod­els), 2) when and how to use dif­fer­ent forms of cen­ter­ing in order to prop­er­ly spec­i­fy mod­els, 3) mod­el bina­ry out­comes, 4) deal with pesky non-con­ver­gence issues, and 5) con­duct mul­ti­level regres­sion diag­nos­tics.  Atten­dees should already have a good grasp of stan­dard regres­sion tech­niques.  Atten­dees maybe those who are new to MLM or those who may already be famil­iar with MLM using oth­er software.

Wednes­day, 8/17, 9:00 am – 4:00 pm, Townsend 222

Meta Analy­sis (Dr. Bixi Zhang)

In this ses­sion, Dr. Zhang will intro­duce meta-analy­sis and how meta-analy­sis can be con­duct­ed in sta­tis­ti­cal com­put­ing soft­ware. The top­ics will cov­er effect sizes, pool­ing effect sizes (fixed effects mod­el and ran­dom effects mod­el), mea­sures of het­ero­gene­ity, sub­group analy­ses, and pow­er analy­sis in meta-analy­sis. Advanced top­ics will also be intro­duced briefly in the ses­sion (e.g., robust vari­ance esti­ma­tion in meta-regres­sion, SEM meta-analy­sis), which are relat­ed to the depen­dent effect size issue. The atten­dees will learn 1) method­olo­gies behind each top­ic; 2) how to use R to do a meta-analy­sis; 3) inter­pre­ta­tions of the results and relat­ed plots; and 4) recent devel­op­ment in meta-analy­sis. The ses­sion is designed for researchers who are inter­est­ed in con­duct­ing meta-analy­ses and learn how to use R lan­guage to ana­lyze their search­ing results of stud­ies with sta­tis­ti­cal mod­els. Famil­iar­i­ty with the R lan­guage will be help­ful but not required.

Thurs­day, 8/18, 9:00 am – 4:00 pm, Townsend 222 

The Basics of Bayesian Sta­tis­tics (Dr. Son­ja Winter)

This ses­sion is designed for researchers who would like to bet­ter under­stand Bayesian sta­tis­tics and who are inter­est­ed in incor­po­rat­ing Bayesian meth­ods into their research prac­tices. Using com­mon sta­tis­ti­cal mod­els (e.g., t‑tests or lin­ear regres­sion), Dr. Win­ter will cov­er 1) the key prin­ci­ples of Bayesian sta­tis­tics, 2) how the Bayesian approach dif­fers from the fre­quen­tist approach (e.g., p‑values and con­fi­dence inter­vals), 3) how to use Bayesian meth­ods to esti­mate para­me­ters and test hypothe­ses, and 4) how to report results from a Bayesian analy­sis. No pre­vi­ous expe­ri­ence with Bayesian meth­ods is required. Atten­dees will be intro­duced to sev­er­al (open-source, free) soft­ware pro­grams. Famil­iar­i­ty with the R lan­guage will be help­ful but not required, as all exam­ples will be based on pre-writ­ten code made avail­able to attendees.

Fri­day, 8/19, 9:00 am – 4:00 pm, Townsend 222 

Non-Gauss­ian Causal Infer­ence (Dr. Wolf­gang Wiedermann)

This ses­sion intro­duces mod­ern causal infer­ence approach­es for both, obser­va­tion­al (non-exper­i­men­tal) and ran­dom­ized con­trolled tri­al data. The first half of the ses­sion focus­es on learn­ing causal mech­a­nisms (i.e., empir­i­cal­ly deriv­ing state­ments con­cern­ing cause and effect) from obser­va­tion­al data alone through intro­duc­ing prin­ci­ples and best prac­tice appli­ca­tions of a recent­ly pro­posed sta­tis­ti­cal frame­work called Direc­tion Depen­dence Analy­sis. In the sec­ond half of the ses­sion, prin­ci­ples of dis­tri­b­u­tion­al (causal) treat­ment effects (DTEs) are intro­duced. DTEs extend aver­age (causal) treat­ment effects (ATEs) to inter­ven­tion effects that man­i­fest in changes beyond means of con­structs and are mod­elled using a dis­tri­b­u­tion­al regres­sion approach called Gen­er­al­ized Addi­tive Mod­els for Loca­tion, Scale, and Shape (GAMLSS). Under the DTE prin­ci­ple, inter­ven­tion effec­tive­ness is allowed to man­i­fest in any fea­tures of the out­come dis­tri­b­u­tion, i.e., changes in vari­ance, skew­ness, kur­to­sis, as well as ceiling/floor effects. GAMLSS mod­el build­ing guide­lines will be pre­sent­ed using real-world data appli­ca­tions. Basic famil­iar­i­ty with the lin­ear regres­sion mod­el and the R sta­tis­ti­cal pro­gram­ming envi­ron­ment are assumed.