| Notes |
Overview
HP wanted to help national ministers of education demonstrate the relationship between education, social investment and economic outcomes. HP had a clear vision for the project but needed a data partner to design the data flow, craft the visualizations and deploy a software interface for the final model.
Enter Yet Analytics. Leveraging the power of Yet’s EIDCC data framework, HP was able to take on the challenge of a predictive approach to econometrics. HP’s goal in this project was to produce two separate but related manifestations of the data model: a databook featuring graphic visualizations and an interactive predictive model. The book was first shared during UN Week in late 2016, and the predictive software debuted at the Education World Forum in London in January 2017.
Case Study: A Predictive Model for Learning, Economic, and Social Indicators 2
Yet Analytics CEO Shelly Blake-Plock at the Education World Forum 2017
Challenge: The LESI Project
HP challenged Yet to use global data to build a predictive model to show what the measurable relationship is between educational and social investments and economic outcomes.
Process: Data
Data sources for the Learning, Economic, and Social Index (LESI) project included decades of global data from the World Bank, the World Economic Forum and the United Nations. The first phase of the project was data discovery.
Many thousands of data indicators were considered. The best were sourced
from across the data sets and were grouped and weighted to create aggregate indicators. In the final tally, the predictive model runs on hundreds of different data points as available on a per-country basis.
The LESI databook presents a static visualization of these weighted and raw
data, organized by country and region. The predictive software running on Yet’s EIDCC platform is dynamic and employs neural networks to identify trends and relationships within and among key values in these data sets. The software allows the user to see hypothetical economic outcomes by adjusting variables in the learning and social data.
Case Study: A Predictive Model for Learning, Economic, and Social Indicators 3
LESI Databook
Output: Databook
The Global LESI databook debuted at UN Week in fall 2016. The book contains profiles on 90+ UN countries and 13 global regions, with visualizations and tabular data representing all of the most important high level indicators from the data model.
The visualizations were designed to give a quick and accessible view into the status of each country. Each profile is broken down by the three main categories – Learning, Economic, and Social – and each country has an aggregate indicator that presents an overview of its scores in each area.
Feature: The Fingerprint
The most complex visualization is the overall indicator.
It needs to represent a lot of data while remaining 59
visually interesting and informative. The reader should
be able to see how each country stacks up relative to
one another across Learning, Economic, and Social
indicators by looking only at this visual ‘fingerprint’. The sunburst design was chosen for its ability to present a large density of information while remaining visually simple. A system of colors – blue for education, purple for economic, orange for social – is used to provide consistency and connect the sunburst data to the
Case Study: A Predictive Model for Learning, Economic, and Social Indicators 4
LESI Interactive Predictive Model powered by Yet’s EIDCC platform
more detailed visualizations and statistics in each country spread. The databook’s visualizations were generated programmatically through Yet’s EIDCC.
Output: Predictive Model
The predictive model trains on and recognizes trends in learning, economic and social factors and can forecast the effect of these on GDP.
The model is customizable to specific indicators, allowing the user to choose from hundreds – and in some cases thousands – of available data points. When the user chooses indicators, the model is automatically re-calculated. In the event of a sparse time series, the model can be set either to interpolate data or to scale
in relation to only the years when data from all indicators is present. The system automatically unifies indicator scale across different value types so data in different units – scaled, percentage, monetary – can all be considered together.
Coefficients from an autoregressive integrated moving average (ARIMA) time series model are coordinated with neural network processes trained on GDP and raw indicator data and are used to generate the predictive model. When a user adjusts the value for an indicator, these new values are calculated with the coefficients to generate the adjusted forecast.
For example, the EIDCC may compute the trends and find that in a given country improvement in the math literacy of females as measured by PISA when combined
Case Study: A Predictive Model for Learning, Economic, and Social Indicators 5
with educational gender parity and a sustained level of investment in internet access will yield a significant percentage outcome in the future growth of GDP per capita. The same factors in a different context may result in a different forecast. The key point is that the model provides a way to test hypotheses in real time.
Perhaps most importantly, the EIDCC’s artificial brain – fueled by the data passing along the neural networks – forecasts the timetable of such investment and provides objective guidance by weighting value in the context of dozens of concurrent social and economic factors.
Feature: The Artificial Brain
Beginning with, but advancing beyond methods
of automated multilinear regression analysis, the artificial brain is comprised of neural networks, each trained to identify trends and relationships within and among key variables.
Individual data records are treated as observations which together comprise layers of information. Training data is passed through these networks thousands, even hundreds of thousands of times, in order to learn the trend and relationship between the presented patterns and the constant – in this case, the individual country’s GDP per capita. The neural networks learn the patterns and relationships signaling the differences in variable values year to year in order to forecast GDP.
The result is an artificial brain purpose-built to assist in the identification and forecasting of return on investment in learning, economic, and social endeavors. Expressly built to take into account the temporality and shifting context of data, the artificial brain and its neural networks can be trained for each and every country –
Case Study: A Predictive Model for Learning, Economic, and Social Indicators 6
LESI Databook
meaning that countries themselves may add their own data in order to attain even more precise and relevant forecasts.
Yet Analytics’ EIDCC artificial brain powers the HP Education Data Command Center.
The interactive analytics and data visualizations of the command center provide:
• A real-time comparison of thousands of data points customizable by country
• The ability to identify and choose key variables to drill down into micro-
components of the time series such as the relationship between technology
spending, social trends, and education strategies
• Hypothesis testing on past and future events
• Real-time computation and visualization of multi-year strategic ROI including
social and cognitive measures
Outcomes
For government leaders, this means the ability to demonstrate the responsible and strategic fiscal rigor of a government; the visionary education reform leadership of government and educational leaders; and the prediction of future payback and time horizons for economic and social outcomes. It makes a clear case for investing in human capital from early childhood through tertiary education.
Drawing from a number of sources and data streams ranging from the internationally comparable to the hyper-local, it renders complex data and statistics
Case Study: A Predictive Model for Learning, Economic, and Social Indicators 7
as elegant and accessible data visualizations. And it proves to international financing organizations that development ROI will be measured and met.
The HP Education Data Command Center, powered by Yet Analytics’ EIDCC artificial brain, provides cross-modal quantitative evidence of return-on-investment of technology spending in education as well as predictive insights to maximize a country’s economic and social outcomes as a result of investments in human capital.
The Product: The HP Education Data Command Center
Based on our collaboration, a new cloud-based software tool is now available: the HP Education Data Command Center, powered by Yet Analytics. Featuring real-time interactive analytics supported by machine learning, the command center allows users to visualize and forecast the effects of education and social spending on economic and educational outcomes. The HP Education Data Command Center
is powered by Yet’s EIDCC artificial brain technology, providing organizations the ability to make data-driven decisions about investments in education and social infrastructure.
The Solution: The Yet EIDCC
Deployable to solve a wide spectrum of human capital data challenges, the EIDCC platform is as flexible as it is powerful. Join Yet Analytics in applying this power to new problem sets, new product development and new solutions.
Contact: hello@yetanalytics.com • 443.256.3673
Case Study: A Predictive Model for Learning, Economic, and Social Indicators 8
 |