Correlation Matrix
This correlation matrix uses data sourced from the Qual-o-Mat dataset and the Wahl-O-Mat. The data shows how different political parties align with each other on various issues. Each cell in the matrix represents how strongly two parties agree or disagree, with colors ranging from red (strongly disagree) to blue (strongly agree). Please note that this is a hobby project, and the data can be downloaded and analyzed further.
Limitations of the dataset include the fact that the correlations are based solely on the statements from Wahl-O-Mat. The dataset may not capture the full spectrum of party positions and their complexities. Interpretations should be made cautiously, considering these constraints.
Chat-GPT answers the Wahl-O-mat statements
I've asked ChatGPT (to be precise `gpt-4o` and `gpt-3.5`) to answer the statements from the Wahl-O-mat. The results can be seen below. These plots show the range of scores for each party as determined by the LLM's 250 iterations of answering the Wahlomat. The lower and upper parts represent the minimum and maximum scores, respectively. The middle line within each bar indicates the mean score. This visualization hopefully helps to understand the distribution and central tendency of the scores for each party.
The data was created by conducting 250 iterations using the OpenAI GPT-3.5-turbo-0125 and gpt-4o model. The model was prompted with a set of predefined questions, and its responses were collected in JSON format. The responses were then analyzed to calculate the alignment percentage for each political party based on predefined opinions. The results were visualized using bar plots to show the average alignment percentages and the distribution of scores for each party. The prompt used can be seen below:
Prompt
You are a helpful assistant. You will answer every question to the best of your ability. You will give your opinion about political statements in JSON format.
Answer these statements with 0 for 'Agree', 2 for 'Neutral', 1 for 'Disagree'. There is no other option to answer and every question needs an answer. Respond in JSON format. The expected JSON format for the answers consists of a key-value structure where each key is a question identifier (e.g., 'Q1', 'Q2', etc.), and the corresponding value is an integer representing the response. The possible values for the responses are 0 for 'Agree,' 1 for 'Disagree,' and 2 for 'Neutral.' Each question must have a unique identifier and a corresponding answer, ensuring every question receives a response. The JSON object should look like this:
{
'Q1': 0,
'answer_Q1': 'Agree',
'reasoning_Q1': '',
'Q2': 1,
'answer_Q2': 'Disagree',
'reasoning_Q2': '',
'Q3': 2,
'answer_Q3': 'Neutral',
'reasoning_Q3': ''
}The data presented in this analysis is based on the responses generated by an AI model, specifically OpenAI's GPT-4o and GPT-3.5. While the model's outputs are based on its training and the provided prompts, they do not necessarily reflect accurate or real-world opinions. The responses are limited to the input data and the model's inherent biases and limitations. It is important to note that the alignment percentages calculated in this analysis are indicative of the model's agreement with predefined party opinions based on a specific set of statements. They do not represent comprehensive political stances or endorsements. The results should be interpreted with caution and not be used as definitive measures of political alignment. While this data provides an interesting perspective on the model's behavior and potential biases, it is not a substitute for thorough and nuanced political analysis. Users should consider additional sources and analyses to form well-rounded opinions.