Downloadable Dyadic Foreign Policy Similarity Data Sets

Users are free to recreate the entire universe of foreign policy similarity scores using the functions available in these data, though that would be a tedious use of time. Toward that end, I have the following data sets available for download in either the qs2 framework (.qs) or the native R serialized data frame (.rds).

Data Set Description	Data Source	Link
Signorino and Ritter’s (1999) S (unweighted)	ATOP Alliances	(.qs, .rds)
Signorino and Ritter’s (1999) S (weighted)	ATOP Alliances	(.qs, .rds)
Benati and Capurri’s (2026) A (unweighted)	ATOP Alliances	(.qs, .rds)
Benati and Capurri’s (2026) A (weighted)	ATOP Alliances	(.qs, .rds)
Cohen’s kappa [κ] and Scott’s pi [π]	ATOP Alliances	(.qs, .rds)
Kendall’s Tau-b [τb]	ATOP Alliances	(.qs, .rds)
Benati and Capurri’s (2026) A (unweighted)	UN Resolutions	(.qs, .rds)
Cohen’s kappa [κ] and Scott’s pi [π]	UN Resolutions	(.qs, .rds)

All data sets are directed dyad-year data all-inclusive of the Correlates of War state system. The temporal domain of the alliance data is 1816-2018. The temporal domain of the UN voting data is 1946-2022, though be mindful of the gap year in 1964.

{peacesciencer}’s add_fpsim() function makes use of a subset of the measures available for download. You can download that here, if you’d like. The download_extdata() function in {peacesciencer} will download this and stick it in that package’s extdata directory in order for the add_fpsim() to work.

A Brief Description of Terminology

I am a product of my time, which privileges shorter, punchier, and somewhat oblique variable names in lieu of the more verbose column names with underscores that have proliferated with the rise of IDEs and chatbots. Toward that end, let me underscore my terminology in these column names and how one can see something like sallybwa to mean “Signorino and Ritter’s (1999) S using binary alliance data, weighted by CINC scores and using absolute distances” and understand that pvotev is a dead giveaway that the data are Scott’s (1955) pi for UN General Assembly votes on their original, valued scale.

s/k/p/a

The first letter of the variable always communicates what the foreign policy similarity measure is. s is Signorino and Ritter’s (1999) S. k is Cohen’s (1960, 1968) (k)appa. p is Scott’s (1955) (p)i. a is Benati and Capurri’s (2026) Alignment Index (A), which is basically a chance-corrected S score.

ally/vote

These communicate the information from which a measure of foreign policy similarity is constructed. ally refers to alliance data generated by the ATOP project, covering 1816-2018. vote refers to UN voting data, covering all UN decisions from 1946 to 2022. Note that these four letters always follow the first letter of the variable (which always communicates what the type of measure is).

u vs. w

These are unique to Signorino and Ritter’s (1999) S measure, which I generate only for the alliance data. u means the measure of S is unweighted by anything and that each alliance tie is equally important in a given dyadic pairing. w means the data are weighted by the target state’s CINC score in a dyadic pairing. I maintain [this is generally ill-advised](here, certainly for anything outside the 19th century, but it did reflect common practice in quantitative IR studies for the longest time. Because Benati and Capurri (2026) are offering what amounts to a chance-corrected S measure, I have weighted and unweighted estimates of A as well.

b vs. v

This concerns measures of foreign policy similarity using alliance ties. Standard practice for the longest time suggested you could construct an ordinal measure of an alliance tie on a scale of 0 to 3. 0 would mean no alliance at all. 1 would be an entente. 2 would be neutrality and/or a non-aggression pact. 3 would be a defense pact. In this reading, the alliance data are fundamentally “valued” (v). Discussion on my blog (here and here) and on {peacesciencer}’s documentation underscore why I’ve never liked this convention and think it drastically misreads what the alliance data actually communicate, but this was the convention I remember learning. If you elect to use the alliance data for a measure of foreign policy similarity (i.e. you need the pre-WW2 reach), I would strongly encourage you to use the measure that treats the alliance as strictly binary (b) (where 0 = no alliance contract and 1 = some form of an alliance contract). Note that the UN voting data are always valued, where 1 = yes, 2 = abstain, 3 = nay.

a vs. s

This concerns Signorino and Ritter’s S and Benati and Capurri’s A because both measure a conceptual “distance”. a means the distances are “absolute” distances where s means the distances are squared. I forget off the top of my head whether Signorino and Ritter’s S measure was intended to be done with absolute distances or whether software packages of the time that included this measure (i.e. EUGene) only offered absolute distances and never belabored this choice to the user. The choice is fundamentally arbitrary though squared distances are much more commonly used in most other types of distance and association metrics. Please be advised again that this should not be confused with the a or s that might appear as the first character of the column.

A More Formal Codebook

Variable	Meaning
aallyvus	A using (v)alued, (u)nweighted alliance (ally) data with (s)quared distances
aallyvua	A using (v)alued, (u)nweighted alliance (ally) data with (a)bsolute distances
aallybus	A using (b)inary, (u)nweighted alliance (ally) data with (s)quared distances
aallybua	A using (b)inary, (u)nweighted alliance (ally) data with (a)bsolute distances
aallyvws	A using (v)alued, (w)eighted alliance (ally) data with (s)quared distances
aallyvwa	A using (v)alued, (w)eighted alliance (ally) data with (a)bsolute distances
aallybws	A using (b)inary, (w)eighted alliance (ally) data with (s)quared distances
aallybwa	A using (b)inary, (w)eighted alliance (ally) data with (a)bsolute distances
pallyv	Scott’s (p)i (v)alued alliance (ally) data
pallyb	Scott’s (p)i (b)inary alliance (ally) data
kallyv	Cohen’s (k)appa using (v)alued alliance (ally) data
kallyb	Cohen’s (k)appa using (b)inary alliance (ally) data
sallyvus	S using (v)alued, (u)nweighted alliance (ally) data with (s)quared distances
sallyvua	S using (v)alued, (u)nweighted alliance (ally) data with (a)bsolute distances
sallybus	S using (b)inary, (u)nweighted alliance (ally) data with (s)quared distances
sallybua	S using (b)inary, (u)nweighted alliance (ally) data with (a)bsolute distances
sallyvws	S using (v)alued, (w)eighted alliance (ally) data with (s)quared distances
sallyvwa	S using (v)alued, (w)eighted alliance (ally) data with (a)bsolute distances
sallybws	S using (b)inary, (w)eighted alliance (ally) data with (s)quared distances
sallybwa	S using (b)inary, (w)eighted alliance (ally) data with (a)bsolute distances
taub	Kendall’s Tau-b on valued alliance data (a legacy measure)
avoteva	A using (v)alued UN voting (vote) data, (a)bsolute distances
avotevs	A using (v)alued UN voting (vote) data, (s)quared distances
pvotev	Scott’s (p)i using (v)alued UN voting (vote) data
kvotev	Cohen’s (k)appa using (v)alued UN voting (vote) data