The partial autocorrelation function or PACF measures the linear dependence between two time series observations at a given lag, controlling for the effect of all the previous lags. It is useful for identifying the direct and indirect relationship between two variables.