This is for all 612 lab sections.

Lab 14: Advanced Material: The Grand Finale

Today is our final and most exciting lab for the year, so buckle up your seat belts as we complete our statistical quest. In our last lab adventure we will navigate our way through the wacky world of crossed and nested designs, vanquish our frightening foe multicollinearity, and dive deep into the depths of our statistical knowledge with Fixed and Random effects models.

General Lab Outline:
-The TAs will be conducting a Module 5 demonstration exactly as we expect you to perform Module 5. -We will teach you new SPSS syntax to run crossed designs with long format data (the same as repeated measures with wide format data).

The datasets with syntax for today’s lab are below:

Datasets:
Crossed Data: Long format
Crossed Data: Wide format
Nested Data
Regression Data
Fixed and Random Effects Data
Additional Fixed and Random Effects Data

Note: BE SURE TO WRITE THE SYNTAX ON YOUR OWN USING THE SYNTAX FILES PROVIDED AS A GUIDE.

Relevant tools:
Online t-statistic calculator

Relevant websites:
Crossed designs in SPSS
Crossed Design SPSS syntax: Wide and long format

Previous Labs

Lab 1: ANOVA Demonstration Module and C & S Practice

Practice Example #1

Context
An educational researcher is interested if college students have the ability to predict and postdict test performance in a classroom context. 100 introductory psychology students participated during a semester long course. Research questions were (1) Can students accurately predict test performance? (2) Does accuracy vary with performance? (3) Does prediction accuracy increase over multiple tests?

Participants
Participants were 100 undergraduate college students enrolled in two sections of an introductory educational psychology course at a mid-south university. The two sections were taught by different instructors, but had the same course requirements. Approximately 90% of the students were psychology majors. The remaining 10% of the students took the course as an elective.

Measures
Students took 3 multiple choice exams. These exams differed in length (first exam was 30 questions, second exam 100 questions, third exam 225 questions). Predictions were made before each exam as the percentage of items students expected to answer correctly. Postdictions were made after each exam as the percentage of items they believed they had gotten correct. Participants were scored on the basis of the overall difference between the prediction/postdiction score and the actual score.

Procedure
To help students (during the actual course) develop their self-assessment skills in the preparation for testing, 1 week prior to taking each of the three exams, students were given optional practice exams that were parallel versions of the actual exams. Students were encouraged to use their performance on the practice exams as a way to identify strengths and weaknesses in their understanding of the material. With each optional practice exam, students were given an answer key with page numbers referencing each question. They were urged to take the optional practice exam without referring to the text or their notes, score it, and then, using their text and notes, try to understand their errors. Students were free to discuss the practice exams with other students or the instructor. One week after these practice exams, students in the course were given the real exams. Immediately before each exam, students recorded on a form attached to the front of the exam their predictions of what percentage of items they expected to get correct and their estimated study times. Immediately after the exams, they recorded their postdictions on a form attached at the end.

Results
Participants were broken up into 4 coded groups (poor, below-average, above-average, and excellent) based on overall test performance. For the first hypothesis, it was found that on average, the students on the whole predicted performance accurately, beta = .14, t(99) = 2.53, p < .05. Additionally, there was an interaction effect in that student in the above-average and excellent groups were strongly able to accurately predict and postdict test performance, while the poor and below-average groups were not. Last, for hypothesis 3, the researchers found that there was no effect whatsoever on accuracy in prediction over time for any individual group or on the whole.

Conclusions by researchers

• Students can in general predict performance
• High achieving students have a superior ability to predict performance.
• Prediction of performance does not improve over multiple testing.
  
  Note: This example was loosely based on the article below.
  
  

Hacker, D., Bol, L., Horgan, D., & Rakow.E. (2000). Test Prediction and Performance in a Classroom Context. Journal of Educational Psychology, 92, 160–170.

A Possible Answer:
(x)O 1 X 1 O 1? (x)O 2 X 2 O 2? (x)O 3 X 3 O 3?

Main Threats: Testing (participants took a practice test and estimation prior to the real one), an interaction affect of selection and testing (the better students would be more likely to engage in the practice effect),multiple treatment interference (multiple vivid treatments were given to the students), instrumentation (the test measurement changed dramatically over time).

Practice Example #2

Context
The National Highway Transportation Safety Administration is interested in understanding the causes of highway crashes in order to develop and refine avoidance countermeasures. One hundred vehicles were outfitted with a data collection suite and provided to participants for their own personal use. The primary goal of this research was to gather naturalistic pre-crash data to understand factors that contribute to car crashes.

Participants
One-hundred drivers in the Northern Virginia/Washington, DC metro area were recruited for this study through flyers placed on vehicles and in the classified section of the newspaper. Drivers received leased vehicles and $125 per month for completing the necessary paperwork. To maximize exposure to near-crash events a larger sample of drivers below the age of 25 and drivers who drove more than an average number of miles made up nearly 37% of the sample population.

Measures and Procedure
One-hundred vehicles were instrumented with a sensor and data collection suite that included: five inner and outer vehicle cameras, a collision notification system, GPS system, Doppler radar antennas, incident pushbutton, and a data acquisition system. Vehicles used in this study were the Toyota Camry (17%), Toyota Corolla (18%), Chevy Cavalier (17%), Chevy Malibu (21%), Ford Taurus (12%), and Ford Explorer (15%). Events were classified into crashes, any contact between the subject vehicle and a fixed object, near crashes, a rapid, severe evasive maneuver, or incidents, conflict requiring and evasive maneuver. Crash severity was also recorded across 4 levels, level 4 being least severe, a non-police reported contact or tire strike, and level 1 being most severe, a police-reported air bag deployment. Subjects drove the vehicles for a 12 to 13 month data collection period in their normal driving area. Originally 109 drivers were included in the study, but because family members and friends drove the vehicles data was collected on 132 vehicles.

Results
In all there were 82 crashes, 761 near-crashes and 8,295 incidents. Of all drivers, 7.5% never experienced an event of any severity; 7.4%, however, experienced many incidents and 3 or 4 crashes. An analysis of eye glance behavior using the in car cameras showed that the majority of sources leading to inattention were from wireless devices, internal distractions, and passengers. Cellular phone conversations were involved in a majority of the crashes (8.7% of all crashes) and near crashes. Another interesting finding was that 93% of rear-end crashes involved inattention to the forward roadway.

Conclusions by researchers

• Wireless devices were the leading contributor to inattention
• Interior distractions were the most frequent source of inattention for crashes
• The most common crash type involved a single vehicle and the most common conflict type involved the subject vehicle and a lead car.
• This study was exploratory in nature and it is essential to determine the exposure rates for each of these types of behaviors
  
  National Highway Traffic Safety Administration. (2006, April). 100-Car Naturalistic Study (DOT HS 810 594). Washington, DC: Neale, V.L., Dingus, T.A., Klauer, S.G., Sudweeks, J., & Goodman, M.

A possible answer:
XOOO…

Threats: history, maturation, selection, instrumentation (sensors on car failing with time), selection x maturation, reactivity, just to name a few.

Practice Example #3

Idea
Subliminally priming words quickly (60 ms or less) leads to “diffuse” activation vs. “specific” activation of the concept primed (above 60 ms). For instance, the word “diligent” if primed at above 60 milliseconds will activate notions related to “hard-working” and “persistent,” whereas when primed at less than 60 ms, as a result of the positive valence generally ascribed to the word, “diligent” will only activate the notion of “good” and things associated with good in memory (thus a much more diffuse prime as compared to the specificity of the longer prime). This is intended to have an influence on how a person’s personality is appraised as the shorter primes (as per the authors) were supposed to lead to more “inferences” that were unwarranted by the prime word (i.e., things related to “good” that don’t follow from the prime “diligent” such as being “nice”)

Participants
51 undergraduates from the Netherlands randomly assigned to one of four conditions (prime exposure length [short, long]; prime valence [good, bad]).

Procedure
Participants were seated in front of a computer and asked to engage in a “parafoveal vigilance task” in which they were to indicate which side of the computer screen a “flash” was administered on. Imbedded in these flashes were first a word presented at either < or > 60 ms followed by a “backward mask” (nonsense words and symbols to cover the word in iconic memory (very short term visual memory; this is intended to make the prime “subliminal”).

The participants completed 10 practice trials to familiarize themselves with the task, thereafter 60 trials of the experimental conditions were completed. All “flashes” were presented in the peripheral visual field of the participant. All words were presented in the same font and size and were intended to activate the notion of either “conceited vs. confident.” The first 40 conditions were primed with neutral words (table, chair, tree). In the final 20 trials of the experimental condition the following words were flashed five times in the positive condition: confident, certain, convinced, and secure. In the negative priming conditions, the following words were each flashed five times: arrogant, conceited, bigheaded, vain.

After the priming task, the participants asked to participate in a “separate” study by a colleague of the primary researcher (whether or not they were allowed to decline or whether any one did was not recorded). The participants read a situational vignette about an individual named Ralph who was acting ambiguously “confident” or “conceited.” Participants were then to rate Ralph’s “personality” on several seven-point Likert scales with the following dimensions confident-conceited (prime-relevant) positive-negative, likable-dislikable, kind-unkind, and friendly-unfriendly (likeability) and thrifty-stingy, normal-plain, sweet-aggressive, polite-crude, and kind-dishonest (prime-irrelevant). (alpha reliability of both scales reported to be approx. .75).

Manipulation checks revealed that the participants reported not knowing they were primed with words and reported that they didn’t believe the “two studies” were related.

Results
Mixed Factorial ANOVA (using “scale” [prime relevant, likeability and prime-irrelevant] as a within-subjects factor was used to analyze the data here. The F tests suggested that i] scale (partial eta squared = .13), ii] prime valence(partial eta squared = .32), iii] scale*prime valence(η² = .12), iv] prime valence*prime length(partial eta squared = .07), & v] scale*prime valence*prime length (partial eta squared = .09) were all significant.

Positive primes always led to more positive and negative to more negative prime-relevant judgments irrespective of length of prime (partial eta squared = .53). Likeability differences only arose in the short prime as opposed to long prime conditions (partial eta squared = .27). The authors were silent on the effects of either on the prime-irrelevant scales.

Conclusions by researchers
Long primes are more specific, short primes lead to more inferences not warranted by the prime (generally in the sense of likeability).

Note: Interested in the study, here’s the cite (based on Study 1):

Stapel, D. A., & Koomen, W. (2005).When less is more: Affective primacy for subliminal priming effects. Personality and Social Psychology Bulletin, 31, 1286–1295.

A Possible Answer:
RXO
RXO
RXO
RXO

Threats: history, reactivity, testing w/ ‘X’ interaction, multiple ‘X’ interference

Lab 2: ANOVA Demonstration Module (advanced) and C & S Practice

Practice Example #4

Idea
Although the more severe manifestations of personality disorder (PD), such as borderline and antisocial, have been gaining empirical and clinical attention, much about the etiology and treatment of these complex psychopathological constellations remains unknown. The authors examined the relationship between ethnicity and treatment utilization by individuals with personality disorders (PDs). Lifetime and prospectively determined rates and amounts of mental health treatments received were compared in over 500 White, African American, and Hispanic participants with PDs in a naturalistic longitudinal study.

Participants
Treatment-seeking or recently treated participants ages 18 to 45 years were recruited from clinical services affiliated with each of four Collaborative Longitudinal Personality Disorders Study (CLPS) sites. The sample for this study consisted of 606 participants who provided data on lifetime treatment history at baseline and 547 participants on whom complete 2-year follow-up data on treatment utilization were obtained

Procedure
Experienced research clinicians, trained to adequate levels of diagnostic reliability, determined PD diagnoses at baseline by using the Diagnostic Interview for DSM–IV Personality Disorders. Participants were assigned to one of four PD diagnostic groups: schizotypal (STPD), borderline (BPD), avoidant (AVPD), or obsessive-compulsive (OCPD). Of those who completed both baseline and 2 years of follow-up, distribution by diagnostic group was the following: STPD, 82 (15.0%); BPD, 165 (30.1%); AVPD, 149 (27.2%); and OCPD, 151 (27.6%). Baseline and follow-along versions of the Longitudinal Interval Follow-Up Evaluation (LIFE; Keller et al., 1987) were used to assess past treatment utilization and 2 years of prospective follow-up at 6-, 12-, and 24-month follow-up intervals. Lifetime data, both amounts and rates of use, were collected at the baseline interview for the following treatment modalities: individual psychotherapy, group psychotherapy, family therapy, self-help groups, day treatment, halfway house, psychiatric hospitalization, and psychotropic medications. For the 2-year prospective period, rates of use and mean amounts received were calculated for individual therapy, group therapy, family therapy, self-help groups, medication consultations, psychiatric hospitalization, and emergency room visits. Medication variables included any psychiatric medication, anxiolytic, hypnotic, mood stabilizer, antipsychotic, and antidepressant.

Results

• Psychosocial Treatment Use: African Americans were less likely to report a history of individual psychotherapy and family therapy. Hispanics were less likely than Whites to report individual psychotherapy, family therapy, self-help, day treatment, and psychiatric hospitalization. Hispanics were less likely than African Americans to report lifetime self-help and day treatment. There were no significant differences in reported group therapy or halfway house treatment between the ethnic groups. The 2-year prospective data showed that African Americans continued to report less likelihood than did Whites of receiving any individual psychotherapy but Hispanics did not. Hispanics remained less likely to receive self-help treatment and psychiatric hospitalization than Whites receive. In addition, Hispanic participants were less likely to receive self-help or hospitalization than were African American participants. During this follow-up period, however, Hispanic participants were more likely to receive group treatment than were either Whites or African Americans.

• Medication Use: At baseline, African Americans and Hispanics were less likely than Whites to report having ever received psychotropic medication. This finding held for anxiolytic, mood stabilizing, and antidepressant medications, all of which both African Americans and Hispanics reported only about one-third to one-half as often as did White participants. There were no differences by ethnic group for lifetime hypnotic, antipsychotic, or anticonvulsant medications. Lifetime psychotropic medication use analyses also yielded two PD diagnostic group by ethnicity interactions: for any psychotropic medications and for antidepressants. White participants with BPD were greater than six times more likely than Hispanic participants with BPD and seven times more likely than African American participants with BPD to receive any psychotropic medication. Similarly, White BPD participants were seven times more likely to receive any lifetime antidepressants compared with Hispanic BPD participants and African American participants with BPD. The prospective data show that African Americans remained less likely than Whites to receive any psychotropic medications over the follow-up period but no difference was found for Hispanics, generally speaking. However, anxiolytic, antidepressant, mood stabilizer, and anticonvulsant medications were used less by Hispanics than Whites, and African American participants were less likely than White participants to use anxiolytics and antidepressants.

• Amounts of Treatment Received: Both African Americans and Hispanics received significantly less individual psychotherapy than did Whites prior to study intake. African Americans and Hispanics received less family therapy over their lifetime prior to baseline, and Hispanics used less self-help groups than did African Americans and Whites and had significantly fewer weeks of psychiatric hospitalization than had Whites. There were no differences among the three groups on group therapy, day treatment, and halfway house use. There was a PD diagnostic group by ethnicity interaction for lifetime weeks of psychiatric hospitalization: White participants with BPD had more weeks of psychiatric hospitalization than did Hispanic and African American participants with BPD. The 2-year prospective data showed that African Americans had fewer individual therapy and medication sessions than had Whites. Hispanics received more family therapy than did African Americans and Whites but had fewer weeks of psychiatric hospitalization than did White participants. A PD diagnostic group by ethnicity interaction for number of medication sessions over 2 years indicated that White BPD participants attended more medication sessions than did the Hispanic and African American BPD participants.

Conclusions by researchers
The nexus of PD, race/ethnicity, and mental health service use remains complex and requires further empirical inquiry. Individuals with PDs are often burdened by interpersonal difficulties and other problems of living that complicate seeking and receiving proper mental health treatments. If we are to improve the prospects of all individuals with PDs, we must seriously consider not only the barriers of psychopathology, but the potential barriers of race and ethnicity as well.

Reference:
Bender, D.S. Skodol, A.E., Dyck, I.R., Markowitz, J.C., Shea, M.T., Yen, S., Sanislow, C.A., Pinto, A., Zanarini, M.C., McGlashan, T.H., Gunderson, J.G., Daversa, M.T., & Grilo, C.M. (2007). Ethnicity and Mental Health Treatment Utilization by Patients With Personality Disorders, Journal of Consulting and Clinical Psychology, 75, 992–999.

Practice Example #5

Idea
Subjective perceptions of personal social status may relate to health beyond the effects of objective socioeconomic status (SES). The authors examined the relationship between subjective social status (SSS) and psychosocial, behavioral, and physical cardiovascular risk factors in middle-aged women. Few studies have examined SSS; SES; and psychosocial, behavioral, and physical risk factors within a single methodological framework. In addition, with the current study, we expand on prior research by investigating the association between SSS and social support and health behaviors—relationships not previously explored. Furthermore, our study is the first to our knowledge to evaluate the association between SSS and AmBP. Finally, no study to date has examined or compared both self-report ladders and their health implications in the same study.

Participants
The current sample was a subset of women from a larger study (N=114) concerning associations among occupation, work characteristics, AmBP, and daily psychosocial experiences. Women self-referred to the study in response to flyers and announcements varied to target a range of SES levels (i.e., professional or white-collar, clerical or administrative support, and service or blue-collar workers). Participants were required to be employed 35 hours or more per week, married or living with a partner, and free from CVD and medications with autonomic effects. The SSS measure was added after study onset, and participants who did not complete this measure (n=20) were excluded from the current article. In addition, 2 women who did not report all indicators of objective SES were excluded, for a final sample size of 92. The study was scheduled for 2 consecutive workdays.

Procedure
Participants were given a battery of questionnaires (that gathered information on, e.g., demographics, health behaviors and history, SSS, activities, and psychosocial assessments) on their first day of participation, with instructions to complete the packet prior to their return to the laboratory the following day.

Results
As expected, U.S. and community SSS were significantly, positively correlated, r(90)= .71, p<.01. There were significant differences in U.S., F(3, 88)=4.63, p<.01, but not community SSS, F(3, 88)=1.84, p<.05, among women in different education groups. Fisher’s least significant difference tests revealed that women in the two highest education groups ranked themselves higher than did women in both of the lower education groups on U.S. SSS. The occupation groups differed significantly for U.S., F(2, 89)=5.35, p<.01, and marginally for community SSS, F(2, 89)=2.72, p<.10; for both ladders, women employed in white-collar work reported higher SSS than did women in blue-collar and clerical work. The income groups differed in ratings of both U.S., F(3, 88)=5.08, p<.01, and community SSS, F(3, 88)=4.05, p<.01. Women reporting an income of $70,000 or more ranked themselves higher on both ladders than did women in the three lower income groups, all p’s<.01. Neither community SSS, F(1, 90)=2.28, p>.05, nor U.S. SSS, F(1, 90)=0.15, p>.05, were related to ethnicity. Likewise, U.S. SSS, r(90)=.10, p>.05, and community SSS, r(90)=.16,p>.05, did not relate to age.

In uncontrolled and controlled analyses, the SSS variables accounted for significant variance in depression, anxiety, pessimism, and stress, with medium to large effect sizes. SSS related significantly to social support in the uncontrolled analysis but with SES controlled. In all cases, higher community SSS tended to be associated with better psychosocial functioning, whereas U.S. SSS did not relate to these outcomes when all variables were in the model. SSS also predicted significant variance in fruit and vegetable consumption (medium effect size), with U.S. SSS positively associated with this outcome. The relationship was slightly attenuated with control for SES. Finally, SSS related to kilocalories expended weekly in leisure activity in both the uncontrolled and the controlled analyses. Both SSS measures tended to be positively associated with this outcome, but neither variable was a significant, independent predictor. SSS did not relate to BMI or smoking. The relationship between SSS and clinic BP and AmBP tended to be slightly stronger after SES was controlled relative to uncontrolled analyses. In controlled analyses, SSS related significantly to clinic SBP, DBP, and AmDBP, with small to medium effect sizes. Further, although the omnibus test for AmSBP was nonsignificant, parameter tests were marginally significant for U.S. SSS and significant for community SSS. In both controlled and uncontrolled analyses, contrary to predictions, higher U.S. SSS predicted higher clinic DBP, whereas community SSS did not relate to DBP. Further, in both analyses, higher U.S. SSS also tended to be associated with higher AmDBP and AmSBP, whereas higher community SSS tended to be associated with lower AmDBP and AmSBP.

Conclusions by researchers
This study contributes to a growing body of research suggesting health effects of SSS beyond the impact of SES. Furthermore, the research suggests that U.S. and community SSS may have distinct implications for cardiovascular risk. In combination with prior research, these results suggest that it would be beneficial to include measures of SSS, in addition to more commonly used objective measures, in future studies.

Reference:
Ghaed. S.G. & Gallo, L.C. (2007). Subjective Social Status, Objective Socioeconomic Status, and Cardiovascular Risk in Women. Health Psychology, 26, 668–674.

Practice Example #6

Context:
An organizational researcher is interested testing if interviews and interview structure has an impact on prediction of performance. 1000 employees in a single retail organization spread out over 2 site locations participated during a 12 month span. Research questions were (1) Can interviews predict job performance? (2) Does prediction vary with the amount of structure to the interviews?

Participants
Participants were 1000 employees in a single retail organization spread out over 2 site locations. The two site locations were a manufacturing/shipping site and a corporate headquarters. The employees were evenly distributed across sites (500 employees at each cite were in the study). The corporate employees possessed slightly advanced age and advanced educational levels compared to the shipping employees.\\

Measures and Procedure

Employees were interviewed and given 2 interviews and interview scores before entering the organization (on a 100 point scale that combined the two interviews, with a higher number being better). The nature of these interviews differed within a site location. The shipping site employees were interviewed with a formally structured interview, while at the corporate headquarters the employees were interviewed with an unstructured interview. 1 week after the initial interview, employees were interviewed again in the same format as the first interview. However, for the unstructured interviews, the interviewer was allowed to freely change the format and questions in the interview if they so desired. The structured interview was not changed at all from the first interview to the second. 6 months later, the employees were scored by their supervisors on a 100 point scale that measured job performance. The interview scores were then standardized and correlated with the standardized job performance scores.

Results
The results indicated that structured interviews and unstructured interviews actually had the same predictive ability in predicting job performance, as both interview structures were correlated .20 with performance, at a statistically significant level. The researchers thus conclude that while interviews can predict job performance at a statistically significant rate, interview structure has no affect on predictive validity of interviews.

Lab 3: Bivariate Regression Demonstration Module and Hand Calculations

Today in lab we will demonstrate a bivariate regression module. The hand calculations shown in class can be found here. We will also be providing some example datasets for you to practice hand calculations of various regression parameters in class. Lastly, you will use SPSS to practice running a bivariate regression analysis on a sample dataset provided to you in class.

Lab 4: Bivariate Regression

Lab 5: Multivariate Regression

SPSS Datasets:
Road Construction Bids
1991 US Social Survey
Employee Data
Cars

Lab 6: Multivariate Regression

SPSS Data Sets and Stories
1.Home Prices
Data
Story
2.US Crime Rates
Data
Story
3.Ice Cream Sales
Data
Story
4.State Public Expenditures
Data
Story
Note: Data is publicly available at http://lib.stat.cmu.edu/DASL/DataArchive.html.

Lab 7: Psychometrics

Data Sets
Hand Calculations Data Set
Data Set 1
Data Set 2
Data Set 3
Data Set 4
Data Set 5
Data Set 6
Data Set 7
Data Set 8
----

Lab 8: Psychometrics II

Data Sets
Hand Calculations Data Set
Data Set 1
Data Set 2
Data Set 3
Data Set 4
Data Set 5
Data Set 6
Data Set 7
Data Set 8

Lab 9: PCA

PCA Data Sets
Sleep
The Brain
2007 Yankees
Hermida Daily Data Grind
----

Lab 10: PCA Part Deux

As you requested we have gathered some examples of PCA applied in the “real world.” So yes, you can publish using PCA and yes PCA is useful in all fields of Psychology. See the examples below:

I/O PCA example
Developmental PCA example
Cognitive Psychology example
Biopsych PCA example
School Psychology PCA example
Clinical PCA example
Personality Assessment PCA example

PCA Data Sets: The Sequel

Below are some data sets for in class SPSS practice. For each data set you will:
1) Run PCA in SPSS
2) Interpret and designate a name or construct for each component (remember variable view, rotations, and plots are your friends)
3) Take turns demonstrating a PCA module for your partner

Dataset 1
Dataset 2 (Note: Crime rate is a dependent variable so do not include it in the PCA. Try regressing crime rate on your principal components.)
Dataset 3
Dataset 4
----

Lab 11: EFA

Listed below are some EFA examples from the literature:

Clinical EFA example
Cognitive EFA example
Social EFA example
School EFA example

Datasets:

Dataset 1
Dataset 2
Dataset 3
Dataset 4
Sleep
The Brain
2007 Yankees
The other dataset 4

Some additional datasets from regression:

Road Construction Bids
1991 US Social Survey
Employee Data
Cars

Lab 12: EFA II

We will be continuing the EFA saga this week. In today’s lab we will be working with a 5 course meal of new datasets of varying interpretability. Conduct both PCA and EFA on each dataset making sure to note the results and specific reasons as to the outcomes you observe. The data used in today’s mod is located here.

Today’s Datasets:

Data set 1: Soup du jour
Data set 2: The appetizer
Data set 3: House salad
Data set 4: Entree
Data set 5: Dessert

Differences Between PCA and EFA:
PCA vs. EFA

Lab 13: Advanced Material

Today we will begin to explore the advanced material in SPSS. We will begin by showing you how to conduct the different types of analysis using SPSS and then you will have time to work through them on your won. Remember, it is important that you not only learn how to conduct each of these analysis but that you can logically connect them to the previous topics in the course.

Today’s Datasets:
1. For these datasets replicate the procedures that we demonstrated in class on your own.
Advanced Material Datasets

2. For this dataset you will need to run a repeated measures ANOVA. Think about why you would run a repeated measures ANOVA, the number of factors in this dataset, and the number of levels for each factor.
Fully crossed dataset

3. This dataset has both a between and within subjects factor. First, fun a mixed design ANOVA. Second, add a new variable to the dataset that is nested within the between subjects variable (old). Then run an ANOVA that includes your new nested variable.
Mixed design dataset

Relevant tools:
Online t-statistic calculator