Amit:
I am a relative newbie to data science, but you raise some provocative issues/questions here: (which has led to more questions than answers that hopefully will spur more dialogue from the group) as you are most likely not alone in navigating this space.
How can we overcome of following challenges in funding the solution of intersectional inequality using Network data science?
Are you referring to the implementation of AI workflows as part of this endeavor? (I recently took the course on Coursera and recommend it, especially business and product specialists, and aspiring data scientists looking to expand knowledge in this area as a way to get started).
1.Quantitative analysis masks the truly dynamic associations between individuals' characteristics and outcomes. There are some methodological challenges with applying a 'inter-categorical' approach to quantitative analysis on intersectional inequalities. The first concerns the categorisation of people into predefined categories This is definitely a concern. Categories would only seem useful if it was tied to some institutional or domain-based affiliation who share similar challenges and goals, but as networks evolve over time, this can shift or become limiting. How do you see probabilistic models addressing this versus deterministic ones? Can multiple imputations and pairwise correlations help address this? As content plays a vital role in resonance across networks, the ability to classify and tag content based on it's attributes could address this as well. (container images, information resources, research/white papers, blog posts across mediums) to create resonance that is correlated and not categorized?
2Because of data constraints, important dimensions of inequality may be overlooked. As data proliferates across systems, dimension reduction is a best practice, but to your point, how does feature selection not contribute to bias or in eliminating important features that would skew the data or limit its potential in some way?
3. In such models the coefficient of an interaction term is not easily understood, and the true relation can also go the opposite way (positive or negative). Are you referring to polarity between intent and outcome of a model? An inverse correlation.
4. Next, the presentation and structure of analysis is important in communicating results from quantitative studies that focus on interactions between features, particularly when transmitting results for audiences less experienced in quantitative research interpretation. How does this change depending on the specific application and context (i.e. healthcare, defense, banking, etc) where objectives may vary significantly whereby the weighted values of qualitative may contribute heavily in consumer-based applications versus those which are network/back-end delivery or platform-based solutions?
5. Quantitative data are not often organized in such a way that intersection analysis is easily facilitated. Again, how does implementation of AI workflows and greater availability of algorithm selection and choices in the model building aid in this area along with usefulness of dashboards (beyond the legacy BI dashboards of quick summaries to a more sophisticated visualization of analytic results in order to communicate discoveries) This may help address these areas of concerns to build more explainable AI.
I am interested to hear others' responses as I share your concerns and am convinced that the answers in solving the most significant challenges we face is through leveraging these advanced capabilities, but addressing these issues is key.
------------------------------
Sara Johnson
Founder
Focus Mobility
------------------------------
Original Message:
Sent: Wed May 12, 2021 03:07 AM
From: Amit Mishra
Subject: How network data science can address social problems of intersectional inequality ?
How can we overcome of following challenges in funding the solution of intersectional inequality using Network data science?
- Quantitative analysis masks the truly dynamic associations between individuals' characteristics and outcomes. There are some methodological challenges with applying a 'inter-categorical' approach to quantitative analysis on intersectional inequalities. The first concerns the categorisation of people into predefined categories.
- Because of data constraints, important dimensions of inequality may be overlooked.
- In such models the coefficient of an interaction term is not easily understood, and the true relation can also go the opposite way (positive or negative).
- Next, the presentation and structure of analysis is important in communicating results from quantitative studies that focus on interactions between features, particularly when transmitting results for audiences less experienced in quantitative research interpretation.
- Quantitative data are not often organized in such a way that intersection analysis is easily facilitated.
------------------------------
Amit Mishra
------------------------------