The RILE index of the manifesto dataset is a popular but controversial estimate of parties’ left–right positions. It has been thoroughly criticized, yet a basic question of its validity has not been addressed. In the current article I argue that for the index to be valid, patterns of association presumed by the logic of the index must be present in the data used to calculate it. I apply canonical correlation analysis (CCA) to test these associations within and between the sets of variables that form the left–right (RILE) index, concluding that for countries which have not experienced a communist past the relationship is present, although substantively weak. More important, however, is that for post-communist countries the required associations in the data are clearly not there. The RILE index is therefore an invalid measure of left–right position for this set of countries.