This paper presents a method for estimating the ideology of political YouTube videos using a combination of correspondence analysis and a language model. The approach begins by analyzing political Reddit subreddits to estimate the ideology of YouTube videos linked to them. Correspondence analysis is used to place these videos in an ideological space, and a language model is trained using these estimated ideologies to predict the ideologies of videos not posted on Reddit. The predicted ideologies are validated against human labels. The method is scalable, fast, and can be easily adjusted to account for changes in the ideological landscape. It is applied to the watch histories of survey respondents to evaluate the prevalence of echo chambers on YouTube and the association between video ideology and viewer engagement. The results show that individuals' watch histories favor ideologically congruent content, and that ideologically extreme videos may receive greater or more favorable engagement. The method provides video-level scores based only on supplied text metadata and is validated against domain knowledge, prior literature, and human-labeled data. The approach is demonstrated to be effective in estimating the ideology of political videos and has potential applications in understanding media diets and ideological extremism.This paper presents a method for estimating the ideology of political YouTube videos using a combination of correspondence analysis and a language model. The approach begins by analyzing political Reddit subreddits to estimate the ideology of YouTube videos linked to them. Correspondence analysis is used to place these videos in an ideological space, and a language model is trained using these estimated ideologies to predict the ideologies of videos not posted on Reddit. The predicted ideologies are validated against human labels. The method is scalable, fast, and can be easily adjusted to account for changes in the ideological landscape. It is applied to the watch histories of survey respondents to evaluate the prevalence of echo chambers on YouTube and the association between video ideology and viewer engagement. The results show that individuals' watch histories favor ideologically congruent content, and that ideologically extreme videos may receive greater or more favorable engagement. The method provides video-level scores based only on supplied text metadata and is validated against domain knowledge, prior literature, and human-labeled data. The approach is demonstrated to be effective in estimating the ideology of political videos and has potential applications in understanding media diets and ideological extremism.