This app applies principal component analysis (PCA) to Supreme Court of the United States (SCOTUS) voting data.

• The plot on the left shows where each justice falls along the first two principal components (labeled "Dim1" and "Dim2", where "Dim" is short for "dimension").  The plot on the right shows where each case falls along these components.  Mouse-hover over a case to see how each justice voted.

• The data come from the Washington Law SCOTUS database.  I used the "justice centered" data, which shows how each justice voted on each case.  The votes are coded as binary; each justice is either in the majority or not in the majority.  There are two options for how to handle concurrences.  By default, all concurring justices are treated as being in the majority.  Optionally, concurring justices can be treated as not in the majority if they express substantial disagreement with the reasoning given in the majority opinion.  For more details on how this judgment was made, see the section of the data codebook on "the vote in the case".

• There are five "terms" that can be plotted, going back to 2010 after Kagan joined the court.  I define each "term" as beginning when a new justice joins the court.  

• Cases decided by fewer than 9 justices are excluded from this analysis. 

• By default, unanimous cases are included in the analysis.  The app gives an option to exclude these.  I have no experience in political science, and no well-informed opinion on whether this should be done.  I have seen research on SCOTUS voting behavior (and voting behavior in other small organizations) in which unanimous votes are excluded, on the grounds that they do not help distinguish voters or blocks of voters from one another.  I am personally hesitant to endorse this, as removing unanimous votes affects the total variance in the 9-dimensional space of justice-centered votes,  as well as the percentage of this variance accounted for in the first two principal components.  But my hesitancy is not informed by any special knowledge of the field; take it with a grain of salt.

• Whether or not  unanimous cases are excluded, the first principal component seems to be quantifying something like ideology.  It tends to separate justices by the party of the president who appointed them (Democrat or Republican).  When unanimous cases are included, the second principal component seems to be quantifying something like "tendency to be in the majority".  Unanimous cases end up on one end; cases with fewer justices in the majority end up on the other.  Excluding unanimous cases makes interpreting the second PC more difficult, and I do not attempt an interpretation here (see the link at the bottom for someone else's proposed interpretation).

• I encourage the user to keep in mind that there is no information about "ideology", nor anything besides individual case votes, being used in this analysis.  Interpreting principal components is a subjective endeavor, even when the interpretations seem obvious.  We may look at the horizontal axis and see "ideology", but formally it is a linear combination of binary votes across 9 justices, chosen so as to maximize data variance while being orthogonal to the 8 remaining PCs.  There is no theory-informed model being fit here, and nothing to tell us the extent to which "noise" (or, if you prefer, "extraneous forces") have influenced the results.  Proceed with caution. 

My inspiration for creating this app came from this Politico piece, along with many years of listening to the excellent Advisory Opinions podcast.