Simpson’s Paradox and Vaccines

Link:https://covidactuaries.org/2021/11/22/simpsons-paradox-and-vaccines/

Graphic:

Excerpt:

So what the chart in the tweet linked above is really showing is that, within the 10-59 age band, the average unvaccinated person is much younger than the average vaccinated person, and therefore has a lower death rate. Any benefit from the vaccines is swamped by the increase in all-cause mortality rates with age.

I have mocked up some illustrative numbers in the table below to hopefully show Simpson’s Paradox in action here. I’ve split the 10-59 age band into 10-29 and 30-59. Within each group the death rate for unvaccinated people is twice as high as for vaccinated people. However, within the combined group this reverses – the vaccinated group have higher death rates on average!

I and others have written to ONS, altering them to the concerns that this data is causing. It appears from a new blog they have released that they are aware of the issue and will use narrower age bands in the next release.

Author(s): Stuart Macdonald

Publication Date: 22 Nov 2021

Publication Site: COVID-19 Actuaries Response Group

Principal Component Analysis

Data Science with Sam

Link:https://www.youtube.com/watch?v=Z6feSjobcBU&ab_channel=DataScienceWithSam

Article: https://sections.soa.org/publication/?m=59905&i=662070&view=articleBrowser&article_id=3687343

Graphic:

Excerpt:

In simple words, PCA is a method of extracting important variables (in the form of components) from a large set of variables available in a data set. PCA is a type of unsupervised linear transformation where we take a dataset with too many variables and untangle the original variables into a smaller set of variables, which we called “principal components.” It is especially useful when dealing with three or higher dimensional data. It enables the analysts to explain the variability of that dataset using fewer variables.

Author(s): Soumava Dey

Publication Date: 14 Nov 2021

Publication Site: Youtube and SOA

Ivermectin: Much More Than You Wanted To Know

Link:https://astralcodexten.substack.com/p/ivermectin-much-more-than-you-wanted

Graphic:

Excerpt:


About ten years ago, when the replication crisis started, we learned a certain set of tools for examining studies.

Check for selection bias. Distrust “adjusting for confounders”. Check for p-hacking and forking paths. Make teams preregister their analyses. Do forest plots to find publication bias. Stop accepting p-values of 0.049. Wait for replications. Trust reviews and meta-analyses, instead of individual small studies.

These were good tools. Having them was infinitely better than not having them. But even in 2014, I was writing about how many bad studies seemed to slip through the cracks even when we pushed this toolbox to its limits. We needed new tools.

I think the methods that Meyerowitz-Katz, Sheldrake, Heathers, Brown, Lawrence and others brought to the limelight this year are some of the new tools we were waiting for.

Part of this new toolset is to check for fraud. About 10 – 15% of the seemingly-good studies on ivermectin ended up extremely suspicious for fraud. Elgazzar, Carvallo, Niaee, Cadegiani, Samaha. There are ways to check for this even when you don’t have the raw data. Like:

The Carlisle-Stouffer-Fisher method: Check some large group of comparisons, usually the Table 1 of an RCT where they compare the demographic characteristics of the control and experimental groups, for reasonable p-values. Real data will have p-values all over the map; one in every ten comparisons will have a p-value of 0.1 or less. Fakers seem bad at this and usually give everything a nice safe p-value like 0.8 or 0.9.

GRIM – make sure means are possible given the number of numbers involved. For example, if a paper reports analyzing 10 patients and finding that 27% of them recovered, something has gone wrong. One possible thing that could have gone wrong is that the data are made up. Another possible thing is that they’re not giving the full story about how many patients dropped out when. But something is wrong.

But having the raw data is much better, and lets you notice if, for example, there are just ten patients who have been copy-pasted over and over again to make a hundred patients. Or if the distribution of values in a certain variable is unrealistic, like the Ariely study where cars drove a number of miles that was perfectly evenly distributed from 0 to 50,000 and then never above 50,000.

Author(s): Scott Alexander

Publication Date: 17 Nov 2021

Publication Site: Astral Codex Ten at substack

Leaders: Stop Confusing Correlation with Causation

Link:https://hbr.org/2021/11/leaders-stop-confusing-correlation-with-causation

Excerpt:

A 2020 Washington Post article examined the correlation between police spending and crime. It concluded that, “A review of spending on state and local police over the past 60 years…shows no correlation nationally between spending and crime rates.” This correlation is misleading. An important driver of police spending is the current level of crime, which creates a chicken and egg scenario. Causal research has, in fact, shown that more police lead to a reduction in crime.

….

Yelp overcame a similar challenge in 2015. A consulting report found that companies that advertised on the platform ended up earning more business through Yelp than those that didn’t advertise on the platform. But here’s the problem: Companies that get more business through Yelp may be more likely to advertise. The former COO and I discussed this challenge and we decided to run a large-scale experiment that gave packages of advertisements to thousands of randomly selected businesses. The key to successfully executing this experiment was determining which factors were driving the correlation. We found that Yelp ads did have a positive effect on sales, and it provided Yelp with new insight into the effect of ads.

Author(s): Michael Luca

Publication Date: 5 Nov 2021

Publication Site: Harvard Business Review

Coffee Chat – “Data & Science”

Link:https://www.youtube.com/watch?v=S5GHsjgSl1o&ab_channel=DataScienceWithSam

Video:

Excerpt:

The inaugural coffee chat of my YouTube channel features two research scholars from scientific community who shared their perspectives on how data plays a crucial role in research area.

By watching this video you will gather information on the following topics:

a) the importance of data in scientific research,

b) valuable insights about the data handling practices in research areas related to molecular biology, genetics, organic chemistry, radiology and biomedical imaging,

c) future of AI and machine learning in scientific research.

Author(s):

Efrosini Tsouko, PhD from Baylor College of Medicine; Mausam Kalita, PhD from Stanford University; Soumava Dey

Publication Date: 26 Sept 2021

Publication Site: Data Science with Sam at YouTube

Emerging Technologies and their Impact on Actuarial Science

Link:https://www.soa.org/resources/research-reports/2021/emerging-technologies-and-their-impact-on-actuarial-science/

Full report: https://www.soa.org/globalassets/assets/files/resources/research-report/2021/2021-emerging-technologies-report.pdf

Graphic:

Excerpt:

Technologies that have reached widespread adoption today:
o Dynamic Collaboration Tools – e.g., Microsoft Teams, Slack, Miro – Most companies are now using this
type of technology. Some are using the different functionalities (e.g., digital whiteboarding, project
management tools, etc.) more fully than others at this time.
• Technologies that are reaching early majority adoption today:
o Business Intelligence Tools (Data Visualization component) – e.g., Tableau, Power BI — Most
respondents have started their journey in using these tools, with many having implemented solutions.
While a few respondents are lagging in its adoption, some companies have scaled applications of this
technology to all actuaries. BI tools will change and accelerate the way actuaries diagnose results,
understand results, and communicate insights to stakeholders.
o ML/AI on structured data – e.g., R, Python – Most respondents have started their journey in using
these techniques, but the level of maturity varies widely. The average maturity is beyond the piloting
phase amongst our respondents. These are used for a wide range of applications in actuarial functions,
including pricing business, modeling demand, performing experience studies, predicting lapses to
support sales and marketing, producing individual claims reserves in P&C, supporting accelerated
underwriting and portfolio scoring on inforce blocks.
o Documentation Generators (Markdown) – e.g., R Markdown, Sphinx – Many respondents have started
using these tools, but maturity level varies widely. The average maturity for those who have started
amongst our respondents is beyond the piloting phase. As the use of R/Python becomes more prolific
amongst actuaries, the ability to simultaneously generate documentation and reports for developed
applications and processes will increase in importance.
o Low-Code ETL and Low-Code Programming — e.g., Alteryx, Azure Data Factory – Amongst respondents
who provided responses, most have started their journey in using these tools, but the level of maturity
varies widely. The average maturity is beyond the piloting phase with our respondents. Low-code ETL
tools will be useful where traditional ETL tools requiring IT support are not sufficient for business
needs (e.g., too difficult to learn quickly for users or reviewers, ad-hoc processes) or where IT is not
able to provision views of data quickly enough.
o Source Control Management – e.g., Git, SVN – A sizeable proportion of the respondents are currently
using these technologies. Amongst these respondents, solutions have already been implemented.
These technologies will become more important in the context of maintaining code quality for
programming-based models and tools such as those developed in R/Python. The value of the
technology will be further enhanced with the adoption of DevOps practices and tools, which blur the
lines between Development and Operations teams to accelerate the deployment of
applications/programs

Author(s):

Nicole Cervi, Deloitte
Arthur da Silva, FSA, ACIA, Deloitte
Paul Downes, FIA, FCIA, Deloitte
Marwah Khalid, Deloitte
Chenyi Liu, Deloitte
Prakash Rajgopal, Deloitte
Jean-Yves Rioux, FSA, CERA, FCIA, Deloitte
Thomas Smith, Deloitte
Yvonne Zhang, FSA, FCIA, Deloitte

Publication Date: SOA

Publication Site: October 2021

Introducing the UK Covid-19 Crowd Forecasting Challenge

Link: https://www.crowdforecastr.org/2021/05/11/uk-challenge/

Twitter thread of results: https://twitter.com/nikosbosse/status/1449043922794188807

Graphic:

Image

Excerpt:

Let’s start with the data. The UK Forecasting Challenge spanned a long period of exponential growth as well as a sudden drop in cases at the end of July 3

Especially this peak was hard to predict and no forecaster really saw it coming. Red: aggregated forecast from different weeks, grey: individual participants. The second picture shows the range for which participants were 50% and 95% confident they would cover the true value

….

So what have we learned? – Human forecasts can be valuable to inform public health policy and can sometimes even beat computer models – Ensembles almost always perform better than individuals – Non-experts can be just as good as experts – recruiting participants is hard

Author(s): Nikos Bosse

Publication Date: Accessed 17 Oct 2021, twitter thread 15 Oct 21

Publication Site: Crowdforecastr

Average Annual Temperature for Select Countries and Global Scale

Link: https://github.com/resource-watch/blog-analysis/tree/master/req_016_facebook_average_surface_temperature

Description:

This file describes analysis that was done by the Resource Watch team for Facebook to be used to display increased temperatures for select countries in their newly launched Climate Science Information Center. The goal of this analysis is to calculate the average monthly and annual temperatures in numerous countries at the national and state/provincial level and globally from 1950 through 2020.

Check out the Climate Science Information Center (CSIC) for up to date information on climate data in your area from trusted sources. And go to Resource Watch to explore over 300 datasets covering topics from food, forests, water, oceans, cities, energy, climate, and society. This analysis was originally performed by Kristine Lister and was QC’d by Weiqi Zhou.

Author: Kristine Lister

Date Accessed: 12 Oct 2021

Location: github

Statistics with Zhuangzi

Link: https://tsangchungshu.medium.com/statistics-with-zhuangzi-b75910c72e50

Graphic:

Excerpt:

The next section is another gratuitous dunk on Confucius, but it’s also a warning about the perils of seeing strict linear relationships where there are none. Not only will you continually be disappointed/frustrated, you won’t know why.

In this story, Laozi suggests that Confucius’ model of a world in which every additional unit of virtue accumulated will receive its corresponding unit of social recognition is clearly not applicable to the age in which they lived.

Moreover, this results in a temptation is to blame others for not living up to your model. Thus, in the years following the 2007 crash, Lehman brothers were apostrophised for their greed, but in reality all they had done was respond as best they could to the incentives that society gave them. If we wanted them to behave less irresponsibly, we should have pushed government to adjust their incentives. They did precisely what we paid them to. If we didn’t want this outcome, we should have anticipated it and paid for something else.

Author(s): Ts’ang Chung-shu

Publication Date: 20 Sept 2021

Publication Site: Medium

Applying Predictive Analytics for Insurance Assumptions—Setting Practical Lessons

Graphic:

Excerpt:

3. Identify pockets of good and poor model performance. Even if you can’t fix it, you can use this info in future UW decisions. I really like one- and two-dimensional views (e.g., age x pension amount) and performance across 50 or 100 largest plans—this is the precision level at which plans are actually quoted. (See Figure 3.)

What size of unexplained A/E residual is satisfactory at pricing segment level? How often will it occur in your future pricing universe? For example, 1-2% residual is probably OK. Ten to 20% in a popular segment likely indicates you have a model specification issue to explore.

Positive residuals mean that actual mortality data is higher than the model predicts (A>E). If the model is used for pricing this case, longevity pricing will be lower than if you had just followed the data, leading to a possible risk of not being competitive. Negative residuals mean A<E, predicted mortality being too high versus historical data, and a possible risk of price being too low.

Author(s): Lenny Shteyman, MAAA, FSA, CFA

Publication Date: September/October 2021

Publication Site: Contingencies

Predictably inaccurate: The prevalence and perils of bad big data

Link: https://www2.deloitte.com/us/en/insights/deloitte-review/issue-21/analytics-bad-data-quality.html

Graphic:

Excerpt:

More than two-thirds of survey respondents stated that the third-party data about them was only 0 to 50 percent correct as a whole. One-third of respondents perceived the information to be 0 to 25 percent correct.

Whether individuals were born in the United States tended to determine whether they were able to locate their data within the data broker’s portal. Of those not born in the United States, 33 percent could not locate their data; conversely, of those born in the United States, only 5 percent had missing information. Further, no respondents born outside the United States and residing in the country for less than three years could locate their data.

The type of data on individuals that was most available was demographic information; the least available was home data. However, even if demographic information was available, it was not all that accurate and was often incomplete, with 59 percent of respondents judging their demographic data to be only 0 to 50 percent correct. Even seemingly easily available data types (such as date of birth, marital status, and number of adults in the household) had wide variances in accuracy.

Author(s): John Lucker, Susan K. Hogan, Trevor Bischoff

Publication Date: 31 July 2017

Publication Site: Deloitte

SYSTEMIC DISCRIMINATION AMONG LARGE U.S. EMPLOYERS

Link: https://eml.berkeley.edu//~crwalters/papers/randres.pdf

Graphic:

Abstract:

We study the results of a massive nationwide correspondence experiment sending more than
83,000 fictitious applications with randomized characteristics to geographically dispersed jobs
posted by 108 of the largest U.S. employers. Distinctively Black names reduce the probability of
employer contact by 2.1 percentage points relative to distinctively white names. The magnitude
of this racial gap in contact rates differs substantially across firms, exhibiting a between-company
standard deviation of 1.9 percentage points. Despite an insignificant average gap in contact rates
between male and female applicants, we find a between-company standard deviation in gender
contact gaps of 2.7 percentage points, revealing that some firms favor male applicants while
others favor women. Company-specific racial contact gaps are temporally and spatially persistent,
and negatively correlated with firm profitability, federal contractor status, and a measure of
recruiting centralization. Discrimination exhibits little geographical dispersion, but two digit
industry explains roughly half of the cross-firm variation in both racial and gender contact gaps.
Contact gaps are highly concentrated in particular companies, with firms in the top quintile of
racial discrimination responsible for nearly half of lost contacts to Black applicants in the
experiment. Controlling false discovery rates to the 5% level, 23 individual companies are found
to discriminate against Black applicants. Our findings establish that systemic illegal
discrimination is concentrated among a select set of large employers, many of which can be
identified with high confidence using large scale inference methods.

Author(s): Patrick M. Kline, Evan K. Rose, and Christopher R. Walters

Publication Date: July 2021, Revised August 2021

Publication Site: NBER Working Papers, also Christopher R. Walters’s own webpages