Unit 4 Correlation and proxy measure Proxy measure

• Slides: 10

Unit 4 Correlation and proxy measure

Proxy measure • Likert scale is a form of self-report data. But what if sometimes you have no direct access to the sample or to the variable? • Use proxy

Problem • The average GPA of APU students is higher than their counterparts at Citrus College. • Some researcher hypothesizes that it is due to the difference in social-economic status (SES). Most APU students are from upper or upper middles classes. Their parents have more resources and therefore APU students outperform CC students.

Problem • Whether the performance gap could be attributed to SES is a big question that awaits further investigation, but some people dispute the assertion that the SES of APU students is higher than that of CC students. • But you have no access to the SES information of their parents. What can you do?

Problem • You need some indirect measurement. • One of the possible proxy measures of SES is the vehicle model.

Assignment • Form a small group of 2 -4 people. • Go to the non-faculty/non-staff parking lots of APU (East and West campuses) and CC. Mark down the automobile models, evaluate their quality (condition), and the degree of luxury (use a 5 -point scale, 1 is the lowest whereas 5 is the highest). • Save the data in Excel and upload the file to Sakai.

Assignment School APU APU APU CU CU CU Make VW Acura Suzuki Acura Honda Toyota VW Pontiac Hyundai Ford Toyota MB Nissan Chevy BMW Honda Toyota Nissan Toyota Model Jetta RSX-5 SXL-7 TSX Accord Corolla Jetta G 6 Sonata Escape RAV 4 E Versa Cobalt 650 I Accord Tacoma Sentra Tacoma Quality Luxury 4 4 3 5 2 3 4 3 5 4 1 3 4 4 5 2 4 5 4 3 4 3 3 3 4 4 3 2 4 3 3 5 3 3 4 3

Different types of correlation coefficient • After obtaining the data of two variables, you need to find their relationship. • Pearson’s r is applicable to measure the association between two continuous-scaled variables. • Spearman is suitable to rank-order/ordinal data; not affected by sample size.

Different types of correlation coefficient • But when you have one binary variable (e. g. “yes/no”, “male/female”) and one continuous-scaled variable, the point biserial correlation (PBC) or biserial correlation (BC) coefficient should be used instead. • In the point biserial correlation coefficient the binary variable should be "naturally" dichotomous, such as gender. • If the binary variable is artificially dichotomized, this variable may be viewed as having an underlying continuity. For example, an instructor may use a cut score to classify the examinees into “pass” and “fail”. In this case, the biserial correlation is the more appropriate approach.

Misconception: small is insignificant • It is important to point out that sometimes even a very low correlation may have clinical significance. For example, Thomas Holmes and Richard Rahe investigated over 5, 000 medical patients to see whether stress might cause sickness. Patients were asked to tally a check list of 43 life events. It was found that there was a positive correlation of 0. 118 between these two variables. Based on this finding, the Holmes-Rahe Stress Scale was developed and validated. 0. 118? What a big deal! But in clinical psychology or medical research, we must pay attention to even such a small correlation. http: //www. harvestenterprisessra. com/The%20 Holmes-Rahe%20 Scale. htm