Insights from Unemployment and Sentiment Data During COVID-19

SitesUSA has created an excellent data visualization tool to help analyze unemployment and sentiment data during COVID-19. Here's a breakdown of the datasets.

Co-authored by Elyse Robinson, Caldwell Smith, and Jimmy Chase.


Data has been a critical component of the commercial real estate industry for years and is becoming a vital differentiator between failure and success. As COVID-19 began to impact the US, we started searching for ways to help our industry and others navigate the uncertainty that lay ahead.

AGS has, for many years, been at the forefront of demographic research and analysis. Led by veterans of the industry, AGS's team was able to create methods for accurately reporting on unemployment numbers with weekly updates.

Spatial.ai's expert team brought new thinking to the equation. By tracking the positive and negative social media reactions across the nation, Spatial.ai is able to portray how communities are responding to the pandemic.

In this article, AGS and Spatial.ai share details on how the data is collected. Read below for more information!

Tracking Unemployment

unemployment over time

About the Data (provided by AGS)

Using the employment by occupation table from the AGS 19B release and unemployment estimates as a baseline, we created a series of simple models by occupation of the speed and severity of expected impacts of the shutdown. Clearly, we anticipate that the first sectors impacted are those of public facing businesses such as retail, restaurants, and hospitality. Over time, other sectors will be increasingly affected even as the curve “flattens” for those first affected sectors. From the updated baseline of 29 February, we have collected the weekly unemployment claims by state. The first two weeks were used as a recent control series, with the third and fourth weeks actually showing the first definitive job losses that can be directly attributed to the COVID-19 shutdowns.

Since those first few weeks, we decided to implement a second level of controls on disaggregating to the block group level. A second layer of vulnerability was added to account for losses by industry by developing a “sector vulnerability” score for each industry, then creating an overall score for each block group. In effect, we estimate the cross tabulation of labor force employment by occupation and employment by industry. This has the effect of modulating losses in high risk occupations which are not necessarily tied to high risk sectors and vice-versa. This has significant impacts, for example, on workers of all occupations in areas with concentrations of food service and hospitality industry.

Along the way, we have modified our methodology as data has become available – unemployment insurance claims do not necessarily translate into unemployed persons – there are many issues with the reporting of individual states which make these models more difficult. In this most recent data release, we have updated our methodology to reflect areas where unemployment rates are decreasing due to the reopening of several states, the removal of individuals from the unemployment count because they no longer seeking employment or have permanently left the labor force, and those who have accepted part-time work, and are no longer considered unemployed. 

AGS’ goal in releasing this data is to show the effects of state and national policies on local areas for policy makers, and to help our primary client base – retail and restaurants – understand this crazy environment and assist in their often survival-oriented business decisions.

Insights

We have seen that the heavy tourist areas (Las Vegas, Orlando, etc.) were hit right at the beginning of the pandemic and continue to suffer. These will also be some of the hardest areas to bounce back, as many Americans will stay home throughout the summer.


At the beginning of the pandemic, there were some surprising impacts on the medical profession, as many doctor and dentist offices were closed, and some remain closed. Also affected in this industry are those in medical and pharmaceutical sales, who would typically go into hospitals and medical offices that are not allowing non-patient visitors inside.

As we look towards the rebound period, communities with diverse job opportunities will bounce back faster than those that have more specialized occupations. We also anticipate that the general “outlook” of the population will have an effect on rebound – some of our data from behavioral surveys shows areas where optimism or pessimism reign, and how those relate to individual characteristics. 

Explore Unemployment Data Viz Tool

Tracking Social Sentiment

social sentiment over time

About the Data (provided by Spatial.ai)

As part of an effort to provide open source resources and data related to the COVID-19 outbreak, this sentiment dataset includes counts of social media posts (Twitter) aggregated at the county level that mention COVID-19. This data is provided historically week over week as far back January 26th, 2020. Posts are classified as negative, neutral, or positive. To perform sentiment analysis, the VADER (Valence Aware Dictionary and sEntiment Reasoner) model was used. 

One of the key benefits of this dataset (particularly for retailers and restaurants) is its ability to signal which communities have more positive sentiment, and are comfortable returning to shopping and dining out. Counties with a generally positive sentiment may be more likely to re-engage with the economy. Pessimistic counties may be less likely to do so.

Insights

We saw a surge in the volume of social posts mentioning COVID-19 the week of March 7th. This is when news of the pandemic spread rapidly across the media and the announcements of stay-at-home orders. The initial spike was the most negative. After peak volume, overall national sentiment moved in a positive direction, likely due to posts encouraging perseverance (#InThisTogether). A closer look made it clear that people tended to be very negative towards the federal government, mixed about state government, and generally positive when it came to the local efforts. After the initial flurry of social posts, we see a steady decline in volume. We also see a convergence of the levels of positive and negative sentiment. Perhaps this indicates an adjustment to the “new normal” after the initial shock of the pandemic.

Just before the massive media surge, we saw rising negative sentiment in heavy tourist areas (Las Vegas, Orlando, etc.) as they anticipated the coming effects of COVID-19. And as AGS mentioned, these areas were indeed some of the hardest hit. Negative sentiment continued to rise the following week but was then quickly overtaken by positive sentiment as people rallied together on social media.

Looking ahead, we expect a continued decline in overall social media volume as businesses reopen and a sense of normalcy returns. In the case of a second wave, it will be especially important to keep an eye on the sentiment of relevant counties, as this can be a leading indicator of that population’s fear related to the virus.

Explore Sentiment Data Viz Tool

Caldwell Smith
Director of Marketing / SitesUSA
Elyse Robinson
Director of Marketing and Biz Dev / AGS
Jimmy Chase
Head of Marketing & Design / Spatial.ai

Want to see the data for yourself?

Download Sample Data