Overview¶

Given COVID relief efforts for the unemployed, there is ongoing debate as to whether stimulus checks and unemployment insurance (UI) are disincentivizing workers from getting back to work. The common belief is that unemployment insurance is causing workers to stay at home during the pandemic -- after all, if you're a low income worker and unemployment benefits pay just as much, if not more than, your normal pay, why wouldn't you stay at home? This belief is restated by several articles over the last year (e.g., AP News, WSJ, WSJ, NYTimes) as well as several companies' management teams (e.g., DPZ).

However, Janet Yellen and the U.S. Department of the Treasury have suggested that, contrary to popular belief, unemployment insurance is not causing workers to stay home during the pandemic:

But, you know, when we look across states or across sectors or across workers — and if it were really the extra benefits that were holding back hiring, you’d expect to see that in — either in states or for workers or in sectors where the replacement rate due to UI is very high, you’d expect to see lower job-finding rates. And in fact, what you see is the exact opposite.

(Source)

That is an extraordinary claim. We're seeing several outlets and companies claim they're having problems with labor shortage as a result of the unemployment insurance despite the high unemployment rates -- and yet, Yellen makes the exact opposite claim.

Let's see if we can corroborate some of these claims.

Fact Checking Yellen: Do Higher UI Benefits Lead to Lower Unemployment?¶

On the surface, this seems very straightforward to fact check: regress on each state's unemployment rate versus its unemployment insurance benefits. However, there are different facets here that will affect our conclusion.

First, how can we define unemployment? Unemployment rates are reported by the Bureau of Labor Statistics every month (here), giving multiple different metrics, both seasonally adjusted and unadjusted:

U-1, persons unemployed 15 weeks or longer, as a percent of the civilian labor force;
U-2, job losers and persons who completed temporary jobs, as a percent of the civilian labor force;
U-3, total unemployed, as a percent of the civilian labor force (this is the definition used for the official unemployment rate);
U-4, total unemployed plus discouraged workers, as a percent of the civilian labor force plus discouraged workers;
U-5, total unemployed, plus discouraged workers, plus all other marginally attached workers, as a percent of the civilian labor force plus all marginally attached workers; and
U-6, total unemployed, plus all marginally attached workers, plus total employed part time for economic reasons, as a percent of the civilian labor force plus all marginally attached workers.

The most commonly reported of these measures is the U-3 unemployment rate, which is what you'll find in most news reports. U-4, U-5, and U-6 rates are also all of interest to us. For our purposes, we prefer the state-level unemployment reporting, which we can find at the Bureau of Labor Statistics' LAU (local area unemployment) here.

Second, how can we define the level of unemployment insurance benefits? There are several problems that we might find in trying to define it. How many people are getting UI? How frequently is the UI distributed? How is UI staggered? Who qualifies for UI? Further, on a per-state basis, not all dollars are equal. The spending power of a dollar in California, for instance, may be significantly different from a dollar in Mississippi -- so while the nominal value of benefits may be the same in two states, for instance, higher spending power in one state may disincentivize workers from getting back into the workforce more. These are all natural complications to defining unemployment insurance benefits. We can dig deeper into states' unemployment insurance benefits by looking at the Consolidated Appropriations Act (here), which provisions Federal Pandemic Unemployment Compensation (FPUC) for each state, which more explicitly define weekly benefits. There are lots of data sources that aggregate each state's weekly benefits.

As an initial pass, we can take a look at states' weekly UI benefits as both averages and maximum (nominal dollar amounts) versus their U-3 and U-6 unemployment rates and see if there is a trendline we can extrapolate.

In [1]:

from collections import defaultdict
import pandas as pd 
import plotly.express as px
import plotly.io as pio

pio.renderers.default = "png"
pio.kaleido.scope.default_width = 1000
pio.kaleido.scope.default_height = 714
df_raw = pd.read_excel(
    "data/unemployment_benefits.xlsx",
    header=1,
    usecols="B:AG",
)

In [2]:

# Massaging the data to create subplots quickly
d = defaultdict(list)
for _, row in df_raw.iterrows():
    d["State"].append(row["State"])
    d["Weekly Benefit"].append(row["Avg Weekly Benefit"])
    d["Weekly Benefit Type"].append("Avg")
    d["Unemployment Rate"].append(row["U-3"])
    d["Unemployment Rate Type"].append("U-3")
    
    d["State"].append(row["State"])
    d["Weekly Benefit"].append(row["Avg Weekly Benefit"])
    d["Weekly Benefit Type"].append("Avg")
    d["Unemployment Rate"].append(row["U-6"])
    d["Unemployment Rate Type"].append("U-6")
    
    d["State"].append(row["State"])
    d["Weekly Benefit"].append(row["Max Weekly Benefit"])
    d["Weekly Benefit Type"].append("Max")
    d["Unemployment Rate"].append(row["U-3"])
    d["Unemployment Rate Type"].append("U-3")
    
    d["State"].append(row["State"])
    d["Weekly Benefit"].append(row["Max Weekly Benefit"])
    d["Weekly Benefit Type"].append("Max")
    d["Unemployment Rate"].append(row["U-6"])
    d["Unemployment Rate Type"].append("U-6")
    
df = pd.DataFrame.from_dict(d, orient="columns")
df["Unemployment Rate"] /= 100.0

fig = px.scatter(
    df,
    x="Weekly Benefit",
    y="Unemployment Rate",
    facet_col="Unemployment Rate Type",
    facet_row="Weekly Benefit Type",
    trendline="ols",
)
fig.update_layout(
    xaxis_tickformat="$",
    xaxis2_tickformat="$",
    yaxis_tickformat="%",
    yaxis3_tickformat="%",
    title="Unemployment Rate vs. Weekly UI by State",
    title_x=0.5,
)
fig.show()

This might be what Yellen alludes to -- you would expect that the higher the weekly benefits in a state, the fewer people would be incentivized to find a job. However, we see that the opposite is true: we find that the higher a state's weekly benefits, the more people actively look for jobs. Note here two things. First, the U-3 rate refers to people who are unemployed and looking for jobs as a percentage of the total civilian labor force, and the U-6 rate includes marginally attached individuals, which are discouraged workers and people who have chosen not to return to the labor force. Second, we are looking at job-seeking rates, and not necessarily job-finding rates, for which we don't have data insights.

This, however, doesn't provide the full picture. There are a few glaring problems with Yellen's conclusion.

First, a dollar in California likely doesn't go as far as a dollar in Alabama. It may be the case that the nominal value of the weekly benefits in Alabama may be lower than the max weekly benefits in California, but after adjusting for cost of living, the benefits in Alabama might actually go a bit further than in California.

Second, and more importantly, the U-3 and U-6 rates do not take into account the actual number of people receiving unemployment insurance benefits. We would expect that the higher the state's UI benefits, the higher the number of claims for UI benefits -- which is what we are seeing. However, whether unemployment claims actually convert to employment is a completely separate question.

Furthermore, if we're concerned about how effective increasing UI benefits is for driving employment, we should be interested in things other than job-seeking rates, such as decreasing the ratio of discouraged workers to unemployed job-seekers, or more comprehensively, looking at U-6 unemployment versus total dollars spent in UI per capita (We would hope to see lower U-6 unemployment rates, which include discouraged workers, as total dollars spent in UI per capita goes up.).

Let's adjust the data to account for geographic variations in the value of the dollar (data) and also factor in the number of individuals actually receiving UI benefits (data).

In [3]:

# Massaging the data to create subplots quickly
d = defaultdict(list)
for _, row in df_raw.iterrows():
    d["State"].append(row["State"])
    d["Adjusted Total Benefits Paid"].append(row["Adjusted Total Benefits Paid"])
    d["Population"].append(row["Population"])
    d["Unemployment Rate"].append(row["U-3"])
    d["Unemployment Rate Type"].append("U-3")
    
    d["State"].append(row["State"])
    d["Adjusted Total Benefits Paid"].append(row["Adjusted Total Benefits Paid"])
    d["Population"].append(row["Population"])
    d["Unemployment Rate"].append(row["U-6"])
    d["Unemployment Rate Type"].append("U-6")
    
df = pd.DataFrame.from_dict(d, orient="columns")
df["Unemployment Rate"] /= 100.0
df["Benefits Paid Per Capita"] = df["Adjusted Total Benefits Paid"] / df["Population"]

fig = px.scatter(
    df,
    x="Benefits Paid Per Capita",
    y="Unemployment Rate",
    facet_col="Unemployment Rate Type",
    trendline="ols",
)
fig.update_layout(
    xaxis_tickformat="$",
    xaxis2_tickformat="$",
    yaxis_tickformat="%",
    yaxis_title="Unemployment Rate",
    title="Unemployment Rate vs. Benefits Paid Per Capita by State",
    title_x=0.5,
)
fig.show()

We are actually seeing a more positive trend between benefits paid per capita and unemployment rate, relative to looking at just weekly benefits. Why could this be the case?

Three possibilities come to mind. First, there could be a high number of individuals actually receiving unemployment insurance benefits. Second, there could be variability in the amounts given out per week (e.g., Benefits could taper off toward the end of the 26-week maximum UI coverage period.). Third, individuals can be extending their UI benefits claims for long periods of time.

Two things can help clarifying what is actually the case:

Investigate the recipiency rate (percentage of unemployed individuals actually receiving UI benefits). We would hope that the weekly benefits offered will not affect the recipiency rate.
Investigate the exhaustion rate (percentage of people receiving UI benefits that go the maximum UI coverage period without actually finding a job). We would hope that weekly benefits offered will not affect exhaustion rate (i.e., people are actually looking for jobs instead of just running it out to completion and collecting the money.).

In [4]:

# Massaging the data to create subplots quickly
d = defaultdict(list)
for _, row in df_raw.iterrows():
    d["State"].append(row["State"])
    d["Weekly Benefit"].append(row["Avg Weekly Benefit"])
    d["Weekly Benefit Type"].append("Avg")
    d["Recipiency Rate"].append(row["Recipiency Rate"])
    
    d["State"].append(row["State"])
    d["Weekly Benefit"].append(row["Max Weekly Benefit"])
    d["Weekly Benefit Type"].append("Max")
    d["Recipiency Rate"].append(row["Recipiency Rate"])
    
df = pd.DataFrame.from_dict(d, orient="columns")
df["Recipiency Rate"] /= 100.0

fig = px.scatter(
    df,
    x="Weekly Benefit",
    y="Recipiency Rate",
    facet_col="Weekly Benefit Type",
    trendline="ols",
)
fig.update_layout(
    xaxis_tickformat="$",
    xaxis2_tickformat="$",
    yaxis_tickformat="%",
    title="Recipiency Rate vs. Weekly Benefits by State",
    title_x=0.5,
)
fig.show()

This is counterintuitive. You would expect that recipiency rates would be unaffected by the amount of weekly benefits offered. But the higher the weekly benefits, the more unemployed people are getting UI benefits. Some possibilities for this may be that the states with higher weekly benefits can afford higher recipiency rates (e.g., maybe there are fewer unemployed people on the whole). Another possibility may be that more people are requesting UI where the weekly benefit amounts are higher. We can check both of these possibilities.

In [5]:

fig = px.scatter(
    df_raw,
    x="Avg Weekly Benefit",
    y="Unemployed Population (000s)",
    trendline="ols",
)
fig.update_layout(
    xaxis_tickformat="$",
    title="Population Unemployed vs. Weekly Benefits by State",
    title_x=0.5,
)
fig.show()

This makes sense -- the number of unemployed people doesn't vary significantly as a function of weekly benefits.

In [6]:

df = df_raw[["State", "Avg Weekly Benefit", "Total Weeks Claimed", "Population"]].copy()
df["Weeks Claimed Per Capita"] = df["Total Weeks Claimed"] / df["Population"]

fig = px.scatter(
    df,
    x="Avg Weekly Benefit",
    y="Weeks Claimed Per Capita",
    trendline="ols",
)
fig.update_layout(
    xaxis_tickformat="$",
    title="Total UI Claims Per Capita vs. Weekly Benefits by State",
    title_x=0.5,
)
fig.show()

This seems a bit extraordinary. The total number of claims goes up as weekly benefits go up, which is exactly what media outlets and companies have called out, and this seems to be what drives high recipiency rates as weekly benefits increase.

The next step is to check out the exhaustion rate, which is the percentage of insured unemployed individuals who collect UI benefits for the maximum amount of time possible.

In [7]:

# Massaging the data to create subplots quickly
d = defaultdict(list)
for _, row in df_raw.iterrows():
    d["State"].append(row["State"])
    d["Weekly Benefit"].append(row["Avg Weekly Benefit"])
    d["Weekly Benefit Type"].append("Avg")
    d["Exhaustion Rate"].append(row["Exhaustion Rate"])
    
    d["State"].append(row["State"])
    d["Weekly Benefit"].append(row["Max Weekly Benefit"])
    d["Weekly Benefit Type"].append("Max")
    d["Exhaustion Rate"].append(row["Exhaustion Rate"])
    
df = pd.DataFrame.from_dict(d, orient="columns")
df["Exhaustion Rate"] /= 100.0

fig = px.scatter(
    df,
    x="Weekly Benefit",
    y="Exhaustion Rate",
    facet_col="Weekly Benefit Type",
    trendline="ols",
)
fig.update_layout(
    xaxis_tickformat="$",
    xaxis2_tickformat="$",
    yaxis_tickformat="%",
    title="Exhaustion Rate vs. Weekly Benefits by State",
    title_x=0.5,
)
fig.show()

This is actually counterfactual to the picture Yellen seems to be promulgating. The higher the weekly UI benefits offered, the more people are claiming benefits to their maximum extent. You would expect that the weekly UI benefits should not affect the percentage of people claiming UI benefits to the end of their coverage period (implying that across all states, people who are collecting unemployment insurance are making similar efforts to find a job). However, that isn't the case. It appears as though the higher the weekly benefits offered, the more people collect UI benefits for as long as possible.

Conclusion¶

Although Yellen's proposition seems to be true that states with higher UI benefits seems to have more people seeking jobs, this is an incomplete picture. We see that when states offer higher weekly benefits, unemployed people make more UI claims per capita and, once they receive UI benefits, exhaust the benefits to the full extent allowed.