Topic models are powerful methods to track the latent semantic structure in large collections of unstructured text. Recently, Roberts et al. (2014) introduced the structural topic model (STM), which makes these models interesting for a broad application in the social sciences. In an improvement to standard topic models, STM promises to control for different sources of bias in the text data as well as to allow experimentation and inferential analyses. One of the most important sources of bias in text collections are time dependencies. In most text collections, serial correlation possibly undermines the dynamic estimation of time-independent concepts such as media attention to policies or issues. In STM, time series can be modelled by using time trends or fixed effects. In this study, it is tested whether this is sufficient to measure the dynamic of media attention on simulated data sets as well as a large corpus of newspaper reports related to smoking bans in the US from 1996 until 2013.