Total: 1
Empirical analyses often compare distributions of ordinal variables across groups or over time using repeated cross-sectional data, where only marginal distributions are observed and missing data are pervasive. As a result, the joint distribution linking these marginals is not identified, making it difficult to assess how observed differences arise. This paper studies how distributional change can be measured and interpreted under such limited information. I show that the $L_1$ distance between cumulative distribution functions admits an optimal transport representation as the minimal reallocation of probability mass across ordered categories. This representation delivers both a scalar measure of discrepancy and a structured description of how distributional change must occur, which I refer to as minimal-mobility configurations. To address missing data, I adopt a partial identification approach and construct sharp bounds on the marginal distributions. These bounds induce identified sets for both the discrepancy measure and the associated minimal-mobility configurations, providing inference that is robust to nonresponse and a transparent basis for assessing sensitivity to missing data. An empirical illustration using data from the \emph{Arab Barometer} demonstrates how the framework can be used in practice to quantify and interpret distributional change under limited information.