“…Quantitatively, prior literature has examined cross‐sectional variation in comment letters based on observable features such as the number of unique topics discussed (Heese et al 2017), the number of filings (Bozanic et al 2019), and the length of the conversation in days or rounds (Cassell et al 2013; Gunny and Hermis 2020). 23 Qualitatively, previous research using textual analysis of comment letters has focused on the length in the number of characters or words (Johnson et al 2020; Lowry et al 2020; Shen and Tan 2020), readability (Ballestero and Schmidt 2019; Cassell et al 2019), negative tone (Agarwal et al 2017; Chantziaras et al 2021; Ege et al 2020), linguistic strength (Liu and Moffitt 2016), parsing of staff information (Ege et al 2020; Kubic 2021), or identification of specific topics (Dechow et al 2016). More sophisticated incorporation of textual analysis includes latent Dirichlet allocation to classify comment letters textually as important or unimportant based on their ability to predict future restatements and write‐downs (Ryans 2021) and Kullback‐Leibler divergence to map textual components of comment letters to changes in future disclosures (Lowry et al 2020).…”