Permalink
Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
536 lines (397 sloc) 19.5 KB
---
title: "'Dangerous Dyads' and an Introduction to Applied Research"
subtitle: POSC 3610 -- International Conflict
author: Steven V. Miller
institute: Department of Political Science
titlegraphic: /Dropbox/teaching/clemson-academic.png
date:
fontsize: 10pt
output:
beamer_presentation:
template: ~/Dropbox/miscelanea/svm-r-markdown-templates/svm-latex-beamer.tex
latex_engine: xelatex
dev: cairo_pdf
fig_caption: true
slide_level: 3
make149: true
mainfont: "Open Sans"
titlefont: "Titillium Web"
---
```{r setup, include=FALSE, cache=F, message=F, warning=F, results="hide"}
knitr::opts_chunk$set(cache=TRUE, warning=F)
knitr::opts_chunk$set(fig.path='figs/')
knitr::opts_chunk$set(cache.path='cache/')
knitr::opts_chunk$set(
fig.process = function(x) {
x2 = sub('-\\d+([.][a-z]+)$', '\\1', x)
if (file.rename(x, x2)) x2 else x
}
)
```
```{r loadstuff, include=FALSE}
knitr::opts_chunk$set(cache=TRUE)
library(tidyverse)
library(stevemisc)
library(knitr)
library(stargazer)
library(dotwhisker)
NDY <- read_csv("~/Dropbox/projects/mid-project/gml-mid-data/2.03/gml-ndy-2.03.csv")
Contird <- read_csv("~/Dropbox/data/cow/contiguity/3.2/contdird.csv")
Alliance <- read_csv("~/Dropbox/data/cow/alliance/4.1/alliance_v4.1_by_dyad_yearly.csv")
NMC <- read_csv("~/Dropbox/data/cow/cinc/NMC_5_0.csv")
Polity <- readxl::read_xls("~/Dropbox/data/polity/p4v2016.xls")
Majors <- read_csv("~/Dropbox/data/cow/states/majors2016.csv")
Contdir <- read_csv("~/Dropbox/data/cow/contiguity/3.2/contdir.csv")
Majors %>%
rowwise() %>%
mutate(year = list(seq(styear, endyear))) %>%
ungroup() %>%
unnest() %>%
arrange(ccode, year) %>%
select(ccode, year) %>%
distinct(ccode, year) %>%
mutate(major = 1) -> Majorscy
Majorscy %>%
rename(ccode1 = ccode,
major1 = major) %>%
left_join(NDY, .) %>%
mutate(major1 = ifelse(is.na(major1), 0, 1)) -> NDY
Majorscy %>%
rename(ccode2 = ccode,
major2 = major) %>%
left_join(NDY, .) %>%
mutate(major2 = ifelse(is.na(major2), 0, 1)) -> NDY
Contird %>%
select(state1no, state2no, year, conttype) %>%
left_join(NDY, ., by=c("ccode1"="state1no", "ccode2"="state2no", "year"="year")) %>%
mutate(conttype = ifelse(is.na(conttype), 0, conttype)) -> NDY
Alliance %>%
select(ccode1, ccode2, year, defense, neutrality, nonaggression, entente) %>%
left_join(NDY, .) %>%
mutate(defense = ifelse(is.na(defense), 0, defense),
neutrality = ifelse(is.na(neutrality), 0, neutrality),
nonaggression = ifelse(is.na(nonaggression), 0, nonaggression),
entente = ifelse(is.na(entente), 0, entente)) -> NDY
NMC %>%
select(ccode:cinc) %>%
rename(ccode1 = ccode,
milex1 = milex,
milper1 = milper,
irst1 = irst,
pec1 = pec,
tpop1 = tpop,
upop1 = upop,
cinc1 = cinc) %>%
left_join(NDY, .) -> NDY
NMC %>%
select(ccode:cinc) %>%
rename(ccode2 = ccode,
milex2 = milex,
milper2 = milper,
irst2 = irst,
pec2 = pec,
tpop2 = tpop,
upop2 = upop,
cinc2 = cinc) %>%
left_join(NDY, .) -> NDY
# Bremer looks at development by reference to CINC indicators where economic share > demographic share
# Similarly: militarization is if its military capabilities > demographic share
NMC %>%
group_by(year) %>%
select(ccode, year, irst:tpop, milper, cinc) %>%
mutate(tpopperc = tpop/sum(tpop),
milperc = milper/sum(milper),
econperc = (irst + pec)/sum(irst + pec),
advanced = ifelse(econperc >= tpopperc, 1, 0),
militarized = ifelse(milperc >= tpopperc, 1, 0)) %>%
select(ccode, year, advanced, militarized) %>%
rename(ccode1 = ccode,
advanced1 = advanced,
militarized1 = militarized) %>%
left_join(NDY, .) -> NDY
NMC %>%
group_by(year) %>%
select(ccode, year, irst:tpop, milper, cinc) %>%
mutate(tpopperc = tpop/sum(tpop),
milperc = milper/sum(milper),
econperc = (irst + pec)/sum(irst + pec),
advanced = ifelse(econperc >= tpopperc, 1, 0),
militarized = ifelse(milperc >= tpopperc, 1, 0)) %>%
select(ccode, year, advanced, militarized) %>%
rename(ccode2 = ccode,
advanced2 = advanced,
militarized2 = militarized) %>%
left_join(NDY, .) -> NDY
Polity %>%
select(ccode, scode, year, polity2) %>%
mutate(ccode = ifelse(ccode == 305 & year < 1919, 300, ccode),
ccode = ifelse(ccode == 347, 345, ccode),
ccode = ifelse(ccode == 364, 365, ccode),
ccode = ifelse(ccode == 769, 770, ccode),
ccode = ifelse(ccode == 818, 816, ccode),
ccode = ifelse(ccode == 529, 530, ccode)) %>%
rename(ccode1 = ccode,
scode1 = scode,
polity21 = polity2) %>%
left_join(NDY, .) -> NDY
Polity %>%
select(ccode, scode, year, polity2) %>%
mutate(ccode = ifelse(ccode == 305 & year < 1919, 300, ccode),
ccode = ifelse(ccode == 347, 345, ccode),
ccode = ifelse(ccode == 364, 365, ccode),
ccode = ifelse(ccode == 769, 770, ccode),
ccode = ifelse(ccode == 818, 816, ccode),
ccode = ifelse(ccode == 529, 530, ccode)) %>%
rename(ccode2 = ccode,
scode2 = scode,
polity22 = polity2) %>%
left_join(NDY, .) -> NDY
NDY %>%
mutate(war = ifelse(fatality == 6, 1, 0),
landcontig = ifelse(conttype == 1, 1, 0),
othercontig = ifelse(conttype > 1, 1, 0),
bothdem = ifelse(polity21 > 5 & polity22 > 5, 1, 0),
relpow = ifelse(cinc1 >= cinc2*3 | cinc1*3 <= cinc2, 1, 0),
major = ifelse(major1 == 1 | major2 == 1, 1, 0),
contig = ifelse(conttype >= 1 & conttype <= 4, 1, 0),
bilat = ifelse(numa + numb == 2, 1, 0),
allied = ifelse(defense == 1 | neutrality == 1 | nonaggression == 1 |
entente == 1, 1, 0),
bothadv = ifelse(advanced1 == 1 & advanced2 == 1, 1, 0),
bothmil = ifelse(militarized1 == 1 & militarized2 == 1, 1, 0),
dyad = as.numeric(paste0("1",sprintf("%03d", ccode1),
sprintf("%03d", ccode2)))) -> NDY
# NDYpy <- sbtscs(NDY, midongoing, year, dyad) %>% tbl_df()
M1 <- glm(midonset ~ contig + bothdem + relpow + major + allied + bothadv + bothmil, data=subset(NDY))
```
# Introduction
### Goal for Today
*Introduce the concept of "dangerous dyads" that shape how we study peace science and get you all reading regression tables.*
### Let's Jump Into It
Most stuff we'll discuss this semester will include lengthy treatment of literature and arguments.
- However, we won't be doing any of that today.
Today, we'll get you acclimated with reading regression tables and understanding what is associated with conflict.
- This "dangerous dyads" framework still shapes how researchers explain broad patterns of inter-state conflict.
# Understanding the Dangerous Dyads
## Unit of Analysis
### Research Design: Unit of Analysis
Dyad-years are the most common units of analysis in this line of peace science.
- *Dyad*: a pairing of any two states (e.g. USA-Canada, India-Pakistan, Cameroon-New Zealand).
- *Year:* This should be intuitive...
Most of what we'll be doing is *non-directed* dyads.
- *Non-directed*: USA-Canada and Canada-USA are the same. Useful for explaining simple onsets.
- *Directed*: USA-Canada and Canada-USA are different observations. Useful for explaining initiating/targeting.
Here's what this looks like in table form.
###
```{r sample-dyad-years-usa-can, eval=T, echo=F}
NDY %>%
select(ccode1, ccode2, year) %>%
head(10) %>%
rename(country1 = ccode1,
country2 = ccode2) %>%
mutate(country1 = "USA",
country2 = "Canada") %>%
kable(., caption="A Simple Table of Ten Dyad Years for the U.S. and Canada")
```
## Dependent Variable
### Research Design: Dependent Variable
MID onsets are some of the most common dependent variables in this line of peace science.
- *Dependent variable*: an outcome we want to explain.
- *MID onset*: the initiation of at least one threat, display, or use of force from one dyad member to another.
- We'll discuss MIDs in greater detail soon, esp. how countries can start new MIDs with other MIDs ongoing.
Here's what this looks like in table form.
###
```{r sample-dyad-years-ind-pak, eval=T, echo=F}
NDY %>%
select(ccode1, ccode2, year, midonset, midongoing) %>%
filter(ccode1 == 750 & ccode2 == 770 & year > 1999) %>%
rename(country1 = ccode1,
country2 = ccode2) %>%
mutate(country1 = "India",
country2 = "Pakistan") %>%
kable(., caption="A Simple Table of 11 Dyad Years for India and Pakistan [Data: GML-MID (v. 2.03)]")
```
## Independent Variables
### Research Design: Independent Variables
We'll focus on seven of Bremer's "factors" for his dangerous dyads.
1. Contiguity
2. Joint democracy
3. Power preponderance
4. Major powers
5. Joint alliance
6. Advanced economies
7. Militarization
We'll talk about how we operationalize these.
### Contiguity
Contiguity is routinely the most robust predictor of MID onset.
- We'll talk about "opportunity", "interactions", and "territory" in a later lecture.
For now, the intuition is states closer to each other are more likely to have a MID.
- The contiguity variable assumes a 1 if both states in the dyad are land-contiguous (e.g. USA-Canada) or are separated by <=150 miles of water (e.g. UK-Netherlands).
###
```{r contiguity-relationships, eval=T, echo=F, fig.height=8.5, fig.width = 14}
Contdir %>%
group_by(conttype) %>%
summarize(sum = n()) %>%
ggplot(.,aes(as.factor(conttype), sum)) +
geom_bar(stat="identity", color="black", fill="#619cff") +
theme_steve_web() +
xlab("Contiguity Type") + ylab("Number of Contiguity Relationships") +
scale_x_discrete(labels=c("Land or River Border",
"<=12 Miles of Water",
"<=24 Miles of Water",
"<=150 Miles of Water",
"<=400 Miles of Water")) +
geom_text(aes(label=sum), vjust=-.5, colour="black",
position=position_dodge(.9), size=4, family="Open Sans") +
labs(title = "Most Contiguity Relationships are Land or River Borders",
subtitle = "It's worth noting we're selecting on states that have one of these relationships. Your typical state pairing in the world isn't anywhere near each other (e.g. Mongolia-Nigeria).",
caption = "Data: Correlates of War Direct Contiguity Data Master File (v. 3.2)")
```
### Joint Democracy
That no two democracies have "ever" fought war against each other is the "closest thing to an empirical law" in all political science.
- Our best data to test this is the Polity data, which measures executive constraints and recruitment.
We'll belabor the reasons why in a few weeks, but for now the intuition is joint democracies have fewer MIDs.
- Conventional cutoff: `polity2` score >= 6 for *both* members of the dyad.
- The joint democracy variable is a 1 if that's true, 0 otherwise.
###
```{r distribution-polity2-scores, eval=T, echo=F, fig.height=8.5, fig.width = 14}
Polity %>%
group_by(polity2) %>%
summarize(sum = n()) %>%
filter(!is.na(polity2)) %>%
ggplot(.,aes(as.factor(polity2), sum)) + theme_steve_web() +
geom_bar(stat="identity", color="black", fill="#619cff") +
geom_text(aes(label=sum), vjust=-.5, colour="black",
position=position_dodge(.9), size=4, family="Open Sans") +
xlab("polity2 Score") + ylab("Number of Country-Year Observations") +
labs(title = "The Distribution of All polity2 Scores for All Countries (1800-2016)",
subtitle = "The distribution is clearly bimodal with two prominent peaks at 10 (maximally democratic) and -7 (stable autocracy)",
caption= "Data: The Polity Project (Center for Systemic Peace)")
```
### Power Preponderance
"Power" has a conspicuous place in IR literature. There's a lot we (think we) know about it.
- We'll have a full week on how we tackle the problem of "power" in IR.
For now, we believe power preponderance *decreases* likelihood of MID onset.
- Measurement: variable = 1 if either side in dyad is three times as powerful (CINC scores) than the other side.
### Major Powers
Major powers seem disproportionately responsible for our MIDs and wars, especially in the 19th century.
- CoW has a data set on the composition of "major powers" from 1816 to the present.
- We'll code a 1 if there's at least one major power in the dyad.
###
```{r major-powers-table, eval=T, echo=F}
Majors %>%
select(stateabb,styear, endyear) -> Majors
cbind(Majors[1:7, ], Majors[8:14, ]) %>%
kable(., caption="The Major Powers [Data: Correlates of War (v. 2016)]")
```
### Joint Alliance
Alliances occupy a special/contentious place in the peace science literature.
- Their presence coincides with our most prominent wars (e.g. WWI).
- Major debate about whether alliances lead to peace or war.
At the dyadic level, our intuition is allies are unlikely to fight *each other*.
- We'll code a 1 if both states in the dyad have some kind of an alliance (i.e. defense pact, non-aggression, entente, neutrality), 0 otherwise.
### "Advanced Economies" and Militarization
Bremer also contends states with advanced economies and those more militarized are important factors. Here's (roughly) how Bremer codes them.
- *Jointly advanced economies*: assumes a 1 if *both* dyad members' share of steel production and energy consumption > share of total population in a given year. Otherwise: 0.
- We'd ideally use GDP data for this but we lack great GDP data before WWII.
- *Jointly militarized states*: assumes a 1 if *both* dyad members' share of military personnel (part of CINC) > share of total population in a given year. Otherwise: 0.
- We'll belabor CINC more in the week on "power" but CINC is an index that prominently measures share of military spending/expenditure.
Data come from Correlates of War's National Military Capabilities.
### Writing It Out
Going forward, don't be surprised if I lump all this information to a single slide. Like this:
- *Unit of analysis*: dyad-year
- *DV*: MID onset [0,1]
- *IVs*: Contiguity, joint democracy, power preponderance, major power in dyad-year, joint alliance, jointly advanced economies, jointly militarized.
## Other Notes
### Modeling Notes
Articles you'll read will also mention some kind of methodological notes. These may include:
- "Cubic splines" and "peace years" for "temporal correlation."
- i.e. you're less likely to have a MID onset if you haven't had one in a while.
- "Clustered" standard errors or "country/dyad fixed effects."
- i.e. some countries (e.g. the U.S.) or dyads (e.g. India-Pakistan) may be peculiar for which conflict is more/less likely.
Don't get too in the weeds with these things. Just note they're there to make our inferences stronger.
### Logistic Regression
Almost every regression you encounter in this class will be a logistic/probit regression.
- i.e. the outcome we want to explain (onset or escalation) is "there" (1) or "not there" (0).
Interpret the model in the context of *positive* or *negative* relationships relative to the "likelihood" of the outcome being present.
- *Negative relationship*: as the independent variable goes up (e.g. joint democracy), the likelihood of a MID goes down.
- *Positive relationship*: as the independent variable goes up (e.g. the dyad is contiguous), the likelihood of a MID goes up.
### Our Expectations of the Seven Factors
| **Factor** | **Expectation** |
|------------|:---------------:|
| Contiguous dyad | + |
| Joint democracy | - |
| Power preponderance | - |
| Major power in the dyad | + |
| Shared alliance | - |
| Jointly advanced economies | - |
| Jointly militarized states | + |
Positive (+) relationships increase the likelihood of a MID onset while negative (-) relationships decrease the likelihood of a MID onset (as the value of the factor/independent variable increases).
## Evaluating a Regression Analysis
### Evaluating a Regression Analysis
I mention there are three things to do when evaluating a regression analysis.
1. Know (however, general) the data used.
2. Know what the objects in the regression table are saying.
3. Know what the regression table *isn't* saying.
### Interpreting the Regression Output
Find the following objects in the regression table:
1. The numbers in parentheses
2. The numbers *not* in parentheses
3. The asterisks that appear next to some numbers.
### Interpreting the Regression Output
1. The numbers in parentheses
- These are the **standard errors**.
- They communicate a prediction error (of sorts).
- However, their interpretation depends on the associated numbers not in parentheses.
2. The numbers *not* in parentheses
- These are the **regression coefficients**.
- They communicate the estimated change in the DV for a unit-change in the IV.
- Determine if it's positive or negative (recall: relationships).
- However, their substantive interpretation depends on the presence/absence of asterisks.
3. The asterisks that appear next to some numbers.
- These communicate the **statistical significance**.
- i.e. is the estimated positive/negative effect discernible from zero?
- If so, we say that the IV has a "significant" (i.e. highly unlikely to be zero) effect on the DV.
## The Dangerous Dyads
###
```{r dangerous-dyads-regression-table, eval=T, echo=F, results="asis"}
stargazer(M1, header=FALSE, font.size="scriptsize",
style="ajps",
dep.var.labels="MID Onset in a Dyad-Year",
title = "A Simple 'Dangerous Dyads' Model of MID Onset for All Dyad-Years, 1816-2010",
covariate.labels = c("Contiguous Dyad", "Joint Democracy",
"Power Preponderance", "Major Power in the Dyad",
"Shared Alliance", "Jointly Advanced Economies",
"Jointly Militarized States"),
omit.stat = c("aic", "ll"),
omit=c("Constant"),
notes = c("Data: GML-MID Data (v. 2.03)"))
```
###
```{r dangerous-dyads-dwplot, eval=T, echo=F, fig.height=8.5, fig.width = 14}
broom::tidy(M1) %>%
# , dot_args = list(size = .75, aes(colour = model, shape = 4))
relabel_predictors(c(contig = "Contiguous Dyad",
bothdem = "Joint Democracy",
relpow = "Power Preponderance",
major = "Major Power in the Dyad",
allied = "Shared Alliance",
bothadv = "Jointly Advanced Economies",
bothmil = "Jointly Militarized States")) -> Mdf
Mdf %>%
dwplot(.,dot_args = list(aes(colour = model, shape = model), size = 2.5),
whisker_args = list(size = 1),by_2sd = TRUE) +
theme_steve_web() +
theme(legend.position = "none") +
xlab("Coefficient Estimate") +
geom_vline(xintercept = 0, colour = "grey60", linetype = 2) +
labs(title = "A Simple 'Dangerous Dyads' Model of MID Onset for All Dyad-Years, 1816-2010",
subtitle = "Contiguity, the presence of a major power, and jointly advanced economies have the largest absolute effects.",
caption = "Conflict data: GML-MID (v. 2.03). Democracy: Polity project (v. 2016). All others: Correlates of War.")
```
# Conclusion
### Conclusion
Bremer's "dangerous dyads" still shape much of what we know of conflict onset.
- Contiguity and major power presence make for greater likelihood of MID onset.
The hope is this lecture at least partially demystified regression tables.
- They'll constitute the bulk of the articles we read, so you should get used to them.