Transcript USING DUMMY VARIABLES

USING DUMMY VARIABLES
OUTLINE
1. Types of relationship: spurious,
enhancement, specification (interactive)
2. Creating “dummy” variables
3. Using dummy variables:
 In enhancement relationships (change of
intercept but not of slope)
 In interactive relationships (change of slope as
well as of intercept)
4. Illustration: Religion, Wealth, and Politics
TYPES OF RELATIONSHIP
• Spurious (relationship between Y and X1 vanishes in
presence of X2)
• Enhancement (inclusion of X2 yields higher R2 than
equation with just X1—i.e., affects strength of
relationship)
• Specification--or “interaction” (inclusion of X2
affects form of relationship between Y and X1)
CREATING “DUMMY”
VARIABLES
• Transform categorical variables into intervalscale variables
• Coding convention: 0, 1
• Binary form for each permutation: e.g., North
(0, 1), South (0, 1), East (0, 1), West (0, 1)
ILLUSTRATION 1:
RELIGION, WEALTH, AND POLITICS
Y = support for Democratic Party (%)
X1 = per capita wealth
X2 = predominant religion (0 = non-Catholic, 1 = Catholic;
0 = non-Jewish, 1 = Jewish; 0 = non-Protestant, 1 = Protestant)
ILLUSTRATION 2:
GOVERNMENTAL PERFORMANCE
AND POLITICAL REGIMES
PROBLEMS AND QUESTIONS
Disenchantment with democracy in Latin America—are
popular judgments fair?
1. How to assess democratic performance in Latin
America?
2. What comparison group for Latin America?
3. How to measure democratic vs. nondemocratic
performance across comparison group?
BASIC HYPOTHESES
• H1: The greater the prevalence of democratic rule,
the greater the provision of material benefits.
• H2: The greater the prevalence of undemocratic
(authoritarian) rule, the greater the provision of
material benefits.
• H3: The prevalence of democratic or undemocratic
rule bears no systematic relationship to the relative
provision of material benefits.
DATA AND VARIABLES
• Universe=Middle income countries—55 in
lower-middle, 41 in upper-middle, N=96
• Time frame=1990-2004
• Units of analysis=country-years
• Dependent variables=
– Annual GDP growth rate (%)
– Infant Mortality (per 1,000)
– Primary School Enrollment (%)
INDEPENDENT VARIABLES
• Region @ World Bank (0, 1)
• Democracy @ Freedom House (0, 1)
• Level of democracy @ Smith, Ziegler:
–
–
–
–
Liberal (0, 1)
Illiberal (0, 1)
Semidemocratic (0, 1)
Nondemocratic (0, 0)
• Note: Use of dummy variables
CONTROL VARIABLES
• Wealth:
– Upper/lower income bracket (0, 1)
– GDP per capita (1990, 2000)
• Structure:
– Manufacturing as % GDP (1990, 2000)
FORMAT FOR REGRESSIONS
Ŷ = a + b1Wealth + b2Structure
+ biRegioni … bkRegionk
Ŷ = a + b1Wealth + b2 Structure
+ b3 Democracy
POLITICAL DETERMINANTS
__Growth__ __Mortality__ __Schooling__
(1,088)
(277)
(369)
Democracy (0, 1)
Nondemocracy (0, 0)
Constant:
R2
.4206
-----
-10.1158*
----
6.0896*
----
3.9562*
84.2483*
77.0707*
.0822
.5336
.1758
*Significant at .001 level or better.
DEMOCRACY BY REGION
__ Growth__ __ Mortality__ __ Schooling__
(1,032)
(266)
(346)
Latin American
& Caribbean
Democracy (0, 1)
.2373
-11.8388*
8.5104*
Other Democracy (0, 1)
.3569
-10.4353*
4.5804*
Nondemocracy (0, 0)
----
Constant:
3.9008*
R2
.0690
*Significant at .01 level or better.
----
----
83.8488*
76.6599*
.5336
.2027
IMPLICATIONS…
#1. Democratic governments in Latin America and
elsewhere have performed at least as well as
authoritarian regimes in promoting economic growth.
#2. Democratic governments in Latin America and
elsewhere have consistently outperformed
authoritarian regimes with regard to infant mortality
and school attendance.
#3. The deeper the democracy, the greater the provision
of fundamental social benefits. [not shown here]
… AND CAVEATS
Analysis of variance around the mean does not
resolve legitimate public concern about the
level of the mean:
Average growth is 3.2% not 8.2%.
Mean infant mortality is 30 per 1,000 not 10.
Average primary school attendance is 89% instead
of 99%.
And so on.