Content Bias and Information Compression

An information presenter faces a physical limit of information transmission. She selects and reports a given number of signal realizations from a large set to maximize the decision maker’s utility. The observed content deviates from the implied substance, as the picture of selected signals looks systematically different from the full fundamental picture. Economic contexts, including prior belief, utility shapes, and payoff relevance, drive the deviation. Apparent reporting biases found in such contents, including slanting to the prior belief or extremes, can be explained by the presenter’s selective coverage to elaborate on the more valuable fundamental realizations against the side of prior or extremes, effectively appearing to generate reports by recalibrating fundamentals. Such biases improve welfare. An asymptotic and analytical mapping from fundamentals to report contents is derived to clarify the interpretation of content data and facilitate their analysis. The model relates to empirical content analysis using frequency-based proxies and can analyze contextual effects on contents.