Chapter 7 – 1 Chapter 12: Measures of Association for Nominal and Ordinal Variables Proportional...

Chapter 7 – 1

Chapter 12:Measures of Association for

Nominal and Ordinal Variables

• Proportional Reduction of Error (PRE)• Degree of Association• For Nominal Variables

– Lambda• For Ordinal Variables

– Gamma• Using Gamma for Dichotomous Variables

Chapter 7 – 2

Take your best guess?

The most common race/ethnicity for U.S. residents (e.g., the mode)!

Now, if we know that this person lives in San Diego, California, would you change your guess?

With quantitative analyses we are generally trying to predict or take our best guess at value of the dependent variable. One way to assess the relationship between two variables is to consider the degree to which the extra information of the independent variable makes your guess better.

If you know nothing else about a person except that he or she lives in United States and I asked you to guess his or her race/ethnicity, what would you guess?

Chapter 7 – 3

Proportional Reduction of Error (PRE)

• PRE—the concept that underlies the definition and interpretation of several measures of association. PRE measures are derived by comparing the errors made in predicting the dependent variable while ignoring the independent variable with errors made when making predictions that use information about the independent variable.

Chapter 7 – 4

Proportional Reduction of Error (PRE)

1

21

E

EEPRE

where: E1 = errors of prediction made when the independent variable is ignoredE2 = errors of prediction made when the prediction is based on the independent variable

Chapter 7 – 5

Two PRE Measures:Lambda & Gamma

Appropriate for…

• Lambda NOMINAL variables

• Gamma ORDINAL & DICHOTOMOUS NOMINAL

variables

Chapter 7 – 6

Lambda• Lambda—An asymmetrical measure of

association suitable for use with nominal variables and may range from 0.0 (meaning the extra information provided by the independent variable does not help prediction) to 1.0 (meaning use of independent variable results in no prediction errors). It provides us with an indication of the strength of an association between the independent and dependent variables.

• A lower value represents a weaker association, while a higher value is indicative of a stronger association

Chapter 7 – 7

Lambda

1E

2E1ELambda

where:E1= Ntotal - Nmode of dependent variable

categories

allforcategoryforemodcategory )NN(2E

Chapter 7 – 8

Example 1: 2000 Vote By Abortion Attitudes

Vote Yes No Row Total

Gore 46 39 85Bush 41 73 114

Total 87 112 199

Abortion Attitudes (for any reason)

Table 7.2 2000 Presidential Vote by Abortion Attitudes

Source: General Social Survey, 2002

Step One—Add percentages to the table to get the data in a format that allows you to clearly assess the nature of the relationship.

Chapter 7 – 9

Example 1


Gore 52.9% 34.8% 42.7% 46 39 85

Bush 47.1% 65.2% 57.3% 41 73 114

Total 100% 100% 100% 87 112 199




Now calculate E1

E1 = Ntotal – Nmode = 199 – 114 = 85

Chapter 7 – 10

Example 1

Vote Yes No Row TotalGore 52.9% 34.8% 42.7%

46 39 85Bush 47.1% 65.2% 57.3%

41 73 114Total 100% 100% 100%

87 112 199




Now calculate E2

E2 = [N(Yes column total) – N(Yes column mode)] +

[N(No column total) – N(No column mode)]

= [87 – 46] + …

Chapter 7 – 11

Example 1:


46 39 85Bush 47.1% 65.2% 57.3%

41 73 114Total 100% 100% 100%

87 112 199

Abortion Attitudes (for any reason)Table 7.2 2000 Presidential Vote by Abortion Attitudes


Now calculate E2



= [87 – 46] +[112 – 73]

Chapter 7 – 12

Example 1


46 39 85Bush 47.1% 65.2% 57.3%

41 73 114Total 100% 100% 100%

87 112 199




Now calculate E2



= [87 – 46] + [112 – 73] = 80

Chapter 7 – 13

Example 1: 2000 Vote By Abortion Attitudes


Gore 52.9% 34.8% 42.7% 46 39 85

Bush 47.1% 65.2% 57.3% 41 73 114

Total 100% 100% 100% 87 112 199




Lambda = [E1– E2] / E1

= [85 – 80] / 85 = .06

Chapter 7 – 14

Example 1


46 39 85Bush 47.1% 65.2% 57.3%

41 73 114Total 100% 100% 100%

87 112 199

Abortion Attitudes (for any reason)Table 7.2 2000 Presidential Vote by Abortion Attitudes


So, we know that six percent of the errors in predicting presidential vote can be reduced by taking into account the abortion attitudes.

Lambda = .06

Chapter 7 – 15

EXAMPLE 2:Victim-Offender Relationship and Type of Crime: 1993

*Source: Kathleen Maguire and Ann L. Pastore, eds., Sourcebook of Criminal Justice Statistics 1994., U.S. Department of Justice, Bureau of Justice Statistics, Washington, D.C.: USGPO, 1995, p. 343.

Type of Crime (X)

Victim-OffenderRelationship (Y)

Rape/sexualassault Robbery Assault Total

Stranger 122,090 930,860 3,992,090 5,045,040

Non-stranger 350,670 231,040 4,272,230 4,853,940

Total 472,760 1,161,900 8,264,320 9,898,980

Step One—Add percentages to the table to get the data in a format that allows you to clearly assess the nature of the relationship.

Chapter 7 – 16

Type of Crime (X)



Stranger 26(122,090)

80(930,860)

48(3,992,090) (5,045,040)

Non-stranger 74(350,670)

20(231,040)

52(4,272,230) (4,853,940)

Total 100%(472,760)

100%(1,161,900)

100%(8,264,320) (9,898,980)

Victim-Offender Relationship & Type of Crime: 1993

Now calculate E1

E1 = Ntotal – Nmode = 9,898,980 – 5,045,040 = 4,835,940

Chapter 7 – 17

Type of Crime (X)




80(930,860)

48(3,992,090) (5,045,040)


20(231,040)

52(4,272,230) (4,853,940)

Total 100%(472,760)

100%(1,161,900)

100%(8,264,320) (9,898,980)

Victim-Offender Relationship & Type of Crime: 1993

Now calculate E2

E2 = [N(rape/sexual assault column total) – N(rape/sexual assault column mode)] +

[N(robbery column total) – N(robbery column mode)] +

[N(assault column total) – N(assault column mode)]

= [472,760 – 350,670] + …

Chapter 7 – 18

Type of Crime (X)

Victim-Offender Relationship (Y)

Rape/sexual assault Robbery Assault Total

Stranger 26 (122,090)

80 (930,860)

48 (3,992,090) (5,045,040)

Non-stranger 74 (350,670)

20 (231,040)

52 (4,272,230) (4,853,940)

Total 100% (472,760)

100% (1,161,900)

100% (8,264,320) (9,898,980)

Victim-Offender Relationship and Type of Crime: 1993

Now calculate E2




= [472,760 – 350,670] +[1,161,900 – 930,860] + …

Chapter 7 – 19

Type of Crime (X)




80(930,860)

48(3,992,090) (5,045,040)


20(231,040)

52(4,272,230) (4,853,940)

Total 100%(472,760)

100%(1,161,900)

100%(8,264,320) (9,898,980)


Now calculate E2




= [472,760 – 350,670] +

[1,161,900 – 930,860] + [8,264,320 – 4,272,230] = 4,345,220

Chapter 7 – 20

Type of Crime (X)




80(930,860)

48(3,992,090) (5,045,040)


20(231,040)

52(4,272,230) (4,853,940)

Total 100%(472,760)

100%(1,161,900)

100%(8,264,320) (9,898,980)


Lambda = [E1– E2] / E1

= [4,835,940 – 4,345,220] / 4,835,940 = .10

So, we know that ten percent of the errors in predicting the relationship between victim and offender (stranger vs. non-stranger;) can be reduced by taking into account the type of crime that was committed.

Chapter 7 – 21

λ : Example

Occupation

Nationality Profession Blue-collar Farmer Total

Russian 10 20 10 40

Ukrainian 15 15 20 50

Byelorussian 20 5 5 30

Total 45 40 35 120

Chapter 7 – 22

λ: E1

• E1

=120-50

=30+40

Nationality Total

Russian 40

Ukrainian 50

Byelorussian 30

Total 120

Chapter 7 – 23

λ: E2

• Profession 45-20=10+15=25

• Blue-collar 40-20=15+ 5=20

• Farmer 35-20=10+ 5=15

• E2 =60

70 60

70.14

Profession

10

15

20

45

Blue-collar

20

15

5

40

Farmer

10

20

5

35

Chapter 7 – 24

Exercise

Occupation

Nationality

Russian Ukrainian Byelorussian Total

Profession 10 15 20 45

Blue-collar 20 15 5 40

Farmer 10 20 5 35

Total 40 50 30 120

Chapter 7 – 25

Asymmetrical Measure of Association

• A measure whose value may vary depending on which variable is considered the independent variable and which the dependent variable.

• Lambda is an asymmetrical measure of association.

Chapter 7 – 26

Symmetrical Measure of Association

• A measure whose value will be the same when either variable is considered the independent variable or the dependent variable.

• Gamma is a symmetrical measure of association…

Chapter 7 – 27

Before Computing GAMMA:

• It is necessary to introduce the concept of paired observations.

• Paired observations – Observations compared in terms of their relative rankings on the independent and dependent variables.

Chapter 7 – 28

Tied Pairs• Same order pair (Ns) – Paired observations

that show a positive association; the member of the pair ranked higher on the independent variable is also ranked higher on the dependent variable.

• Ns, are ordered the same on each variable, cluster around the main diagonal, and indicate a positive relationship.

Chapter 7 – 29

Chapter 7 – 30

Tied Pairs• Disagreement pair (Nd) – Paired

observations that show a negative association; the member of the pair ranked higher on the independent variable is ranked lower on the dependent variable.

• Nd, are ordered higher on one variable than the other, cluster about the off diagonal, and suggest a inverse relationship.

Chapter 7 – 31

Chapter 7 – 32

Gamma—a symmetrical measure of association suitable for use with ordinal variables or with dichotomous nominal variables. It can vary from 0.0 (meaning the extra information provided by the independent variable does not help prediction) to 1.0 (meaning use of independent variable results in no prediction errors) and provides us with an indication of the strength and direction of the association between the variables. When there are more Ns pairs, gamma will be positive; when there are more Nd pairs, gamma will be negative.

Gamma

Chapter 7 – 33

Gamma

NdNs

NdNsGamma

Chapter 7 – 34

Example• Fund-raising participation vs. income

Participation

Income

High Medium Low

High 5 1 1

Medium 2 4 2

Low 0 1 4

Chapter 7 – 35

5

4 2

1 4

5 1 1

2 4 2

0 1 4

1

2

4

4

4

5 1 1

2 4 2

0 1 4

5 1 1

2 4 2

0 1 4

5 1 1

2 4 2

0 1 4

2

1 4

Chapter 7 – 36

Ns

5*(4+2+1+4)=5*11=55

1*(2+4)=1*(6)= 6

2*(1+4)=2*(5)=10

4*(4)= 4*(4)=16

Ns =87

5

4 2

1 4

1

2

4

4

4

2

1 4

Chapter 7 – 37

1

2 4

0 1

5 1 1

2 4 2

0 1 4

1

2

0

4

0

5 1 1

2 4 2

0 1 4

5 1 1

2 4 2

0 1 4

5 1 1

2 4 2

0 1 4

2

0 1

Chapter 7 – 38

1

2 4

0 1

1

2

0

4

0

2

0 1

1*(2+4+0+1)=1*7= 7

1*(2+0)=1*(2)= 2

2*(0+1)=2*(1)= 2

4*(0)= 4*(0)= 0

Nd =11

Nd

Chapter 7 – 39

Goodman and Kruskal’s γ

• Gamma 87 11

.7887 11

s d

s d

N N

N N

Chapter 7 – 40

Interpreting Gamma

The sign depends on the way the variables are coded:

+ the two “high” values are associated, as are the two “lows”

– the “highs” are associated with the “lows”

.00 to .24 “no relationship”

.25 to .49 “weak relationship”

.50 to .74 “moderate relationship”

.75 to 1.00 “strong relationship”

NdNs

NdNsGamma

Chapter 7 – 41

• Measures of association—a single summarizing number that reflects the strength of the relationship. This statistic shows the magnitude and/or direction of a relationship between variables.

• Magnitude—the closer to the absolute value of 1, the stronger the association. If the measure equals 0, there is no relationship between the two variables.

• Direction—the sign on the measure indicates if the relationship is positive or negative. In a positive relationship, when one variable is high, so is the other. In a negative relationship, when one variable is high, the other is low.

Measures of Association

Chapter 7 – 1 Chapter 12: Measures of Association for Nominal and Ordinal Variables Proportional...

Documents

Transcript of Chapter 7 – 1 Chapter 12: Measures of Association for Nominal and Ordinal Variables Proportional...