One year into the Lee Jae-myung government, The Kyunghyang Shinmun and Underscore modeled the political positions of all 319 members of the 22nd National Assembly. Instead of relying only on yes-or-no floor votes, we used AI to score four kinds of documents: bills each member sponsored, remarks in committee minutes, press-conference statements at the Assembly press center, and remarks quoted in news coverage. A Bayesian model then combined the scores. The result covers each member's ideology, partisanship, attitude toward the president, and how their position shifted over two years. (For the method, see the methodology note .)
Scores are estimates based on documents from the analysis period. Where intervals overlap, the order of members is uncertain. All figures are shown to three decimal places. A credible interval is the range that contains a member's true position with 94% probability under the Bayesian estimate. The dot marks the estimated mean and the line through it marks the interval. A narrow interval means the data were sufficient for a firm estimate. When two members' intervals overlap, we cannot say which of them is more progressive or more conservative.
Each dot is one member (posterior mean). The horizontal and vertical lines are the 94% credible intervals on each axis. Hover to see the name, coordinates and intervals. Click the legend to show or hide a party.
Each dot is one member. Only the members at the two extremes are labeled. Pink dots mark a critical direction and red dots a favorable one. A gray dot means the 94% credible interval includes zero, so the direction cannot be determined. Horizontal lines are 94% credible intervals. The estimate covers People Power members. Of the party's 112 members, those with fewer than three news items that could be scored for attitude toward Yoon were left out of the model: Baek Jong-heon and Kim Ki-woong had two such items each, and Lee So-hee had none.
Monthly average of the attitude score toward President Lee (−3 strongly hostile to +3 strongly favorable) in news items where a Democratic Party member is the speaker (left axis). The number above each point is the count of valid attitude observations that month. The bars behind show the share of negative scores (−3 to −1) among them (right axis). Neutral (0) counts in the denominator; missing values are excluded. These are raw counts, not model output.
The z-score takes the partisanship θ from the model and computes (θ − party mean) ÷ party standard deviation within each party. Signs are flipped for the Democratic Party and the Rebuilding Korea Party; People Power keeps the original sign. The vote measure is a raw count; the speech measure is the combined estimate standardized within the party.