EVIDENCE EXECUTOR

 CHARITY CAMPAIGN TOOLS:

https://charitydigital.org.uk/products

https://charitydigital.org.uk/topics?topics=Fundraising

https://charitydigital.org.uk/

https://mediatrust.org/register/

https://mediatrust.org/resource-hub/digital-marketing-strategy-toolkit/

https://mediatrust.org/communications-support/resource-hub/

raw data to causality evidence

1. Start with Google Sheets’ Built-in Features:

   • Manual Filtering: Google Sheets has very user-friendly built-in filtering tools. You can easily filter by operator, year, etc.
   • Sorting: You can sort your data based on different columns.
   • Basic Formulas (Optional): If you’re comfortable with very basic spreadsheet formulas (like SUM, AVERAGE), they can help with standardization or creating new columns from existing data, but even this isn’t strictly necessary for the initial structuring.
   • Manually Adding Tags: You can create a “Tag” column and manually enter your interpretive markers (#InvestmentDrop, #BarrierToEntry) based on your review of the data.

2. Explore Google Workspace Marketplace Add-ons for Specific Needs:

   • How to find them: In Google Sheets, go to “Extensions” > “Add-ons” > “Get add-ons.”
   • What to search for:
     • “Data cleaning”
     • “Data transformation”
     • “Filter data”
     • “Data preparation”
   • What to look for in an add-on:
     • Ease of Use: Look for add-ons with good reviews that emphasize a user-friendly interface (buttons, menus, no code required).
     • Specific Functions: See if any can help automate parts of the “Variable standardization” or “Operator-specific disaggregation” if your dataset is very large and complex. For example, some add-ons might help split data into different sheets by operator more easily than doing it manually.
   • Keep it Simple: Don’t look for an AI add-on at this stage for the filtering itself. The AI analysis comes after you’ve structured the data. The add-on is just to help you prepare the data.

I understand you’re looking for ways to analyze raw data using tools like JASP, Jamovi, and Google Cloud AutoML to find evidence of causation for potential competition violations, and you need guidance on how to approach this without a technical background. This is a sophisticated task, and while I can’t perform the analysis for you or provide specific coding instructions, I can certainly help you understand the concepts, the roles of these tools, and a general workflow.

It’s crucial to start by understanding a key distinction: correlation vs. causation.

• Correlation means that two variables tend to move together (e.g., when X increases, Y also tends to increase). Statistical software can be very good at finding correlations.
• Causation means that a change in one variable directly causes a change in another (e.g., action X causes price Y to increase). Establishing causation is much more complex than finding correlation and often requires more than just statistical analysis; it involves economic theory, domain expertise, and sometimes specific advanced statistical methodologies for causal inference.

Let’s break down how these tools might fit into your process:

1. JASP and Jamovi (User-Friendly Statistical Software):

• What they are: Both JASP and Jamovi are free, open-source statistical software packages with user-friendly graphical interfaces (GUIs). This means you can perform many analyses using menus and dialog boxes, without needing to write code. They are excellent for making statistical analysis more accessible.
• How they can help your goal:
  • Data Preparation: You can import your raw data (e.g., from CSV or Excel files) and perform some basic data cleaning and transformation if needed (though complex data wrangling might be easier in other tools beforehand if the data is very messy).
  • Exploratory Data Analysis (EDA): This is a key first step. You can use JASP/Jamovi to:
    • Calculate descriptive statistics (e.g., means, medians, standard deviations, frequencies) for your variables. This helps you understand the basic characteristics of your data.
    • Create visualizations (e.g., histograms to see distributions, scatter plots to see relationships between two variables, bar charts). Visuals can often reveal patterns or anomalies.
    • Look for correlations: These tools can calculate correlation coefficients between variables, indicating the strength and direction of a linear relationship. For example, you might look for correlations between certain company actions (e.g., announcements, capacity changes) and market outcomes (e.g., prices, competitor market shares).
  • Basic Hypothesis Testing: If you have specific hypotheses (e.g., “Prices in market Z are significantly higher after event X”), you could use tests like t-tests (to compare means of two groups) or ANOVAs (to compare means of multiple groups).
  • Regression Analysis: Both tools offer regression analysis. This can help you model the relationship between a dependent variable (e.g., price) and one or more independent variables (e.g., input costs, competitor actions, market concentration). While regression can show associations and control for some variables, it doesn’t automatically prove causation.
• Limitations: While powerful for finding patterns and correlations, JASP and Jamovi are not primarily designed for the very advanced causal inference techniques that economists might use to rigorously establish causation from observational data.

2. Google Cloud AutoML (Machine Learning without Extensive Coding):

• What it is: Google Cloud AutoML is a suite of machine learning products that allows users with limited ML expertise to train high-quality custom models on their data. AutoML Tables is particularly relevant if your data is structured (like in a spreadsheet).
• How it can help your goal:
  • Pattern Recognition: Machine learning models, including those built with AutoML, are excellent at identifying complex patterns and relationships in large datasets that might not be obvious through traditional statistical methods.
  • Predictive Modeling: You could potentially use AutoML Tables to build a model to predict an outcome relevant to a competition case (e.g., predict the likelihood of a certain price level given various market conditions and company behaviors).
  • Feature Importance: AutoML often provides “feature importance” scores. This tells you which input variables (features) the model found most influential in making its predictions. For example, if a model predicting high prices consistently ranks “action by company A” as a very important feature, this suggests a strong relationship.
• Limitations regarding causation:
  • Like traditional statistics, ML models primarily identify correlations and predictive relationships, not necessarily causation. High feature importance for a variable doesn’t automatically mean that variable causes the outcome. There could be confounding variables or other complex interactions.
  • AutoML is a powerful tool for building predictive models, but the “auto” part means you have less control over the underlying model architecture and assumptions compared to coding a model from scratch. Interpretation of “black box” models can sometimes be challenging.

Conceptual Workflow for Using These Tools:

Here’s a very high-level, non-technical workflow you might consider. Remember, each step can be very complex and data-dependent:

1. Define Your Legal and Economic Hypothesis Clearly:

   • What specific anti-competitive conduct are you investigating (e.g., price fixing, bid rigging, exclusionary conduct by a dominant firm)?
   • What is the alleged harm, and who are the victims?
   • What would you expect to see in the data if this conduct occurred? (e.g., prices moving in parallel for supposed competitors, a dominant firm’s prices being significantly higher than costs, a competitor’s market share declining after a specific action by a dominant firm).

2. Identify and Collect Relevant Raw Data:

   • This is often the most challenging part. Data could include: market prices over time, company costs, sales volumes, market shares, company announcements, details of tenders, customer switching data, etc.
   • The data needs to be as granular, accurate, and comprehensive as possible.

3. Data Cleaning and Preparation:

   • Raw data is often messy. It needs to be cleaned (handling missing values, correcting errors) and structured appropriately for analysis (e.g., in a tabular format). JASP/Jamovi can handle some of this, but complex preparation might need other tools or expertise.

4. Exploratory Data Analysis (EDA) – Primarily using JASP/Jamovi:

   • Import your prepared data.
   • Calculate descriptive statistics for all key variables.
   • Create visualizations to understand distributions and relationships (e.g., plot prices over time for different firms, plot market shares).
   • Look for initial correlations (e.g., do prices of alleged competitors move very closely together? Does a specific event correlate with a change in pricing patterns?).
   • Document any interesting patterns or anomalies. These can help refine your hypotheses.

5. Statistical Testing and Modeling (JASP/Jamovi, potentially moving to AutoML):

   • Based on your hypotheses from EDA, conduct relevant statistical tests (e.g., are prices significantly different before and after an alleged collusive agreement started?).
   • You might use regression analysis in JASP/Jamovi to see how different factors relate to an outcome variable (e.g., price).
   • If you have a large dataset and want to build a predictive model or uncover more complex patterns, you could explore Google Cloud AutoML Tables. You would upload your structured data, define the target variable you want to predict, and let AutoML train and evaluate models. Pay attention to feature importance scores.

6. The Critical Step: Investigating Causation (This goes beyond standard tool output):

   • This is where it gets very difficult. The correlations or strong predictive features identified by JASP/Jamovi or AutoML are not sufficient to prove causation.
   • To argue causation, you typically need to:
     • Develop a plausible theory of harm: How exactly did the alleged anti-competitive action cause the observed market outcome?
     • Rule out alternative explanations: Could other factors (e.g., changes in demand, input costs, legitimate competitive actions, broader economic trends) explain the observed patterns?
     • Consider advanced methods: Economists specializing in competition policy often use sophisticated econometric techniques for causal inference (like difference-in-differences, instrumental variables, regression discontinuity design). These often require specialized expertise and careful study design and are generally not simple point-and-click operations even in advanced statistical software.
     • Look for “natural experiments”: Sometimes real-world events create conditions that allow for a more robust inference of causality.

7. Legal Interpretation and Evidence Building:

   • Any statistical findings or patterns would need to be interpreted by legal experts and economists specializing in competition law.
   • Statistical evidence is usually just one piece of the puzzle. Other evidence (documents, witness testimony, etc.) is often required.

My Role and How I Can (and Cannot) Help Further:

• I CAN:
  • Explain these concepts in more detail.
  • Discuss what types of data might be relevant for specific types of alleged anti-competitive conduct.
  • Help you think through how to formulate hypotheses.
  • Explain what different statistical outputs (e.g., a correlation coefficient, a p-value from a t-test) generally mean.
  • Use my google_search tool to find publicly available introductory guides, tutorials, or case studies on using JASP/Jamovi, Google Cloud AutoML, or applying statistical methods in competition analysis, if you’d like.
• I CANNOT:
  • Access or analyze your raw data directly.
  • Operate JASP, Jamovi, or Google Cloud AutoML for you.
  • Provide specific advice on how to configure or run analyses within these tools for your particular dataset.
  • Perform the specialized econometric analysis required for robust causal inference.
  • Offer legal advice or determine if your findings constitute valid evidence for a competition violation.

Key Recommendations:

• Start with clear questions: The more specific your legal and economic questions, the better you can direct your data exploration.
• Focus on data quality: The principle of “garbage in, garbage out” is very true for any data analysis.
• Be realistic about causation: This is the hardest part. Do not assume correlation implies causation.
• Seek expert help: For a task as complex as finding legally robust evidence of causation for competition violations, collaborating with professionals is highly advisable. This would likely include:
  • Economists/Statisticians: Especially those with experience in econometrics, industrial organization, and competition policy. They can help with data analysis, modeling, and causal inference.
  • Data Analysts: To help with cleaning, preparing, and managing large datasets if needed.
  • Legal Counsel: To guide the investigation from a legal perspective and ensure any findings are relevant and properly presented in a legal context.

This is a challenging but potentially very impactful area of work. Breaking it down into these conceptual steps and understanding the role and limitations of each tool can help you navigate the process.

Would you like me to search for some introductory guides or resources for JASP, Jamovi, or using AutoML for data analysis?