OVERVIEW

GUIDE 1: INTRODUCTION GUIDE 2: CONSTRUCTING A TABLE GUIDE 3: UNIVARIATE STATISTICS AND DISPLAYS GUIDE 4: BIVARIATE BASICS GUIDE 5: BIVARIATE CORRELATIONS GUIDE 6: MULTIVARIATE CROSSTABULATIONS GUIDE 7: BASIC REGRESSION GUIDE 8: REGRESSION SPECIFICS GUIDE 9: SAMPLING

TO EDF 5400 READINGS AND ASSIGNMENTS   ASSIGNMENT 5 SPECIFICATIONS COMING SOON

We do not go over PERSONAL papers during class or break. However, I will discuss them after class, and we can discuss them during office hours or through an appointment.

Maria and I are at least as interested in how you arrive at your answer as what your answer is.

This assignment is a good example. Let's suppose you chose the correlation coefficient phi (Cramer's V) to examine the association between degree and study year.

You received partial credit if you did. You COULD use V because you can use this correlation coefficient with any kind of data. But tau-beta was the BETTER choice in this case because (a) year is interval and degree is ordinal and (b) with only two categories in the independent variable (year), you can't tell if the relationship is nonlinear, so you may as well go with a higher level correlation coefficient.

However, if you DID choose Cramer's V, we wanted to make sure you were consistent, and that you subsequently chose the correct numeric values and strengths that belonged with V and not some other correlation coefficient. This became important because the correlation coefficient between year and degree was a weak strength for men if you used tau-beta but a moderate strength if you used V (both were very close in magnitude and the tau was just a tad below moderate level).
 

Despite the "AQ" (Anxiety Quotient) on this exercise, most people did quite well. The median score was 19/20, the mean was 18.2 points (s = 2.5), and the IQR was from 18-20. Really can't do better than that! Especially since all these measures should be very familiar by now.

I know it sounds trite to say live with some anxiety; however, most of us get nervous when we learn new material. There is that ghastly feeling of not quite having one's feet on the floor. But, as you know by this time, such a feeling dissipates with practice. This will happen with basic regression, too.

   If you scored below 16 on this assignment, you are IN TROUBLE and need some extra help. Maria has been absolutely terrific working one-on-one with students (thank you, Maria!) and has office hours in the LRC Tuesday and Thursday 3:30-5:15.
 
 

• Completed all parts of the assignment. CRITICAL!
• Turned in ALL computer output.
• Turned the assignment in ON TIME.

You understood that you could not determine the form of the correlation between sex and degree because sex is (1) a nominal variable and (2) only has two categories or values (NOT "VARIABLES") and form cannot be determined under these circumstances.

You correctly reported the numeric value of the  correlation ( = .065 or .07) and identified its value as VERY WEAK.

You could use Phi  (you lost 1 point if you did) because you can use Phi with any level of data, or with nonlinear relationships. It's just not the BEST correlation under these circumstances.

No one identified Gamma as the best coefficient. Thank you!

You used the CHI-SQUARE value and its accompanying probability level (p < .01) to ascertain that the relationship was real. You did not use the sample value of Tau-b (or Phi) to ascertain if the sample relationship was real or a sampling accident.

You recognized that you could not identify the form of the relationship because you only had two categories on the independent variable, year. You need at least three categories (or VALUES) to identify form.

When you examined the correlations (Phi or Tau-B) separately for women and men, you noted that:

• they were statistically significant in BOTH groups
• the probability levels were p < .01 for both Men and Women
• close to the same value (tau-beta was 0.25 for men and 0.32 for women)--certainly these correlations were within 0.10 of one another.
• weak in magnitude for men and moderate for women (it was moderate for men ONLY if you used Phi)
• almost the same as the tau-beta for everyone combined (0.28)

All three correlations between degree and year were easily within 0.10 of one another, for everyone, for women and for men.

PLUS sex very weakly (but definitely) predicted degree.

• You could say that there is a JOINT RELATIONSHIP among degree level, gender and year. Gender DOES predict  degree level, although very weakly and year certainly predicts degree. Further, the correlation between degree level and year remains roughly constant when gender is controlled.
• OR, you could say that this is an EXTRANEOUS (direct) relationship. Year predicts degree level and the relationship remains roughly constant when gender is controlled. The effect of gender on degree is so weak that gender is a poor control variable in this case.

It is not appropriate to say that one subtable correlation is bigger (e.g., for women) than the other because you do NOT have an interaction effect. Basically the correlation between year and degree is about the same for both sexes.

You kept confusing variables (an entity that varies) with values (the scores or categories that a variable takes on).

You said you could determine the form of the relationship (monotonic, linear or nonlinear) with a nominal variable or a variable that had only two categories or values.

You thought Chi-Square was a correlation coefficient. Or, you thought a correlation coefficient was an inference measure.

Level of data is the first thing we examine when choosing a statistic (for example, you couldn't use Pearson's r here because degree was ordinal. You couldn't use eta--a few people tried--because the dependent variable degreet was ordinal and eta requires a NUMERIC dependent variable.)

You did not recognize the joint effects in question 15. Remember, in an extraneous relationship, the control variable is not related (or is related but the effect is so trivial that it is VERY weak--were it weak instead, this would definitely have been a JOINT relationship).

You never mentioned the relationshp between the control variable gender and degree in deciding between a joint and extraneous relationship. This correlation is what makes the difference.

You said the 0.25 tau-beta between year and degree for men was moderate. It is not, it is weak. Please review THE CHART.
 

POSITIVE POINTS

Almost everyone has become a real "pro" at disentangling statistical significance from the strength of a correlation.

Everyone could locate the appropriate output for the total sample and for each gender subgroup.

Almost everyone could correctly identify the strength of their chosen correlation coefficients.

Most students recognized that there was a joint (or extraneous) effect among gender, degree level, and year and could tell us why.
 
 

PLEASE STUDY  YOUR ASSIGNMENT. COMMENTS ARE ON THEM AS APPROPRIATE.  There will be a comparable problem on Exam Three

READINGS AND ASSIGNMENTS

OVERVIEW

Susan Carol Losh November 15 2004
This page was built with Netscape Composer
and is best viewed with Netscape Navigator
600 X 800 display resolution.