In October 2011, the consulting firm Sentier Research, led by former U.S. Census Bureau officials, began publishing a monthly series of household income data for the United States based on Current Population Survey (CPS) microdata.
For the first time, this new series introduced monthly estimates of U.S. household income, derived from the pretax money income that households report earning in the previous twelve months at the time of the survey. In contrast, the Census household income estimates are reported on a calendar-year basis and suffers from a significant delay between data collection and publication, with data for a given calendar year being published in September of the following year.
Even though the monthly CPS is not designed to focus on income characteristics, the income data included in the survey allows for the computation of a valid monthly estimate of annual household income. Sentier proved the robustness of the new household income series, showing a high correlation between their estimates and the official estimates.
The novel household income estimates were widely cited by national media outlets and policymakers as an important monthly indicator of the economic well-being of households in the United States. Sheila Bair, who served as the Chair of FDIC during the Great Recession, notably chose a chart displaying Sentier’s median household income index for The Washington Post as the most significant graph of 2013 (see article). The Sentier series led Bair to conclude in December 2013 that “for a large number of American households, there has been no economic recovery.”
The Sentier income series kept providing valuable information on the monthly trend of U.S. household incomes in the aftermath of the Great Recession and during the ensuing economic recovery. In February 2020, Sentier announced the cessation of its operations due to financial difficulties, resulting in the discontinuation of the series, which had last been published in December 2019. Since then, a monthly series of U.S. household income exclusively derived from CPS microdata has not been released
Motio Research is now offering a new series of monthly U.S. household income data, following the core methodology employed by Sentier and incorporating improvements that enhance the series’ stability and predictive performance.
Our monthly series of household income data are derived from monthly Current Population Survey (CPS) microdata. The monthly CPS is the primary source of labor force statistics for the United States. Conducted by the U.S. Census Bureau, the CPS is a nationally representative survey with a sample size of roughly 60,000 eligible households. The CPS sample covers the noninstitutionalized civilian population.
The official series of median household income is calculated using a different data set, an annual supplement to the CPS called the Annual Social and Economic Supplement (ASEC), previously known as the March Supplement. This survey has a current sample size of more than 75,000 households and is designed to capture poverty and income characteristics of the previous calendar year, instead of focusing on monthly labor force characteristics.
The income question included in the monthly CPS asks households to report their total income in the last twelve months, including “money from jobs, net income from business, farm or rent, pensions, dividends, interest, social security payments, and any other money income received” from all household members ages fifteen or older. Respondents are asked to select among the following sixteen income intervals or brackets:
Interval | Range |
---|---|
1 | Less than $5,000 |
2 | $5,000 to $7,499 |
3 | $7,500 to $9,999 |
4 | $10,000 to $12,499 |
5 | $12,500 to $14,999 |
6 | $15,000 to $19,999 |
7 | $20,000 to $24,999 |
8 | $25,000 to $29,999 |
9 | $30,000 to $34,999 |
10 | $35,000 to $39,999 |
11 | $40,000 to $49,999 |
12 | $50,000 to $59,999 |
13 | $60,000 to $74,999 |
14 | $75,000 to $99,999 |
15 | $100,000 to $149,999 |
16 | $150,000 or More |
Households included in the CPS are surveyed for 8 months in total, spending 4 consecutive months in the survey since their entrance, being then taken out from the survey for 8 consecutive months, and then being reintroduced into the survey for their final 4 consecutive months. Respondents can be grouped according to their month in the survey, so there are 8 groups of roughly similar size of respondents each month.
A key issue for the computation of our monthly household income estimates is that we only get “fresh” household income data in the first and fifth months of the respondent’s tenure in the survey, with the remaining two sets of consecutive months (2, 3, and 4; and 6, 7, and 8) carrying forward the values reported in months 1 and 5, respectively.
The fresh household income data represents roughly one-fourth of the monthly sample (specifically, two months—1 and 5—out of a total of 8 months). We have however opted to follow Sentier’s decision to use the full sample, taking into account the high volatility of a series solely based on the most current data.
Our monthly household income estimates are thus defined as a four-month moving average measure. The fresh, current month data is represented by households in months 1 and 5 in the survey. One-month-old data is represented by households in months 2 and 6; two-month-old data by those in months 3 and 7; and three-month-old data by those in months 4 and 8.
Imputation of missing data. Since January 2010, the U.S. Census Bureau has been imputing missing data from non-responses in the income question. We have chosen to concentrate on the period with missing data imputed by the Census, starting from January 2010. This decision ensures a more consistent series over time by avoiding a mix of different imputation methods.
1. Derivation of actual income values from income brackets
The first step in the computation of a median household income using the monthly CPS is the imputation of actual income values derived from the sixteen income brackets detailed in the previous section.
For the bottom fifteen income intervals, we used the method of linear interpolation used by Sentier, where income values are derived by income bracket following a uniform distribution.
For the highest bracket, we use a Pareto interpolation, where income values are derived by income bracket following a standard Pareto distribution. This is a superior imputation solution for the upper tail section of income distributions.
2. Inflation adjustment
The household income values derived from the income brackets are expressed in current dollars (nominal terms) and need to be adjusted for inflation to express the estimates in constant dollars (real terms).
The inflation adjustment must be performed at the microdata level due to the co-existence of four different months’ worth of past twelve-months income data within each monthly survey (see the Data section, above). In other words, real median household income values cannot be directly derived from nominal median household income values, and vice versa.
We adjust the household income values using the Chained Consumer Price Index for All Urban Consumers (CCPI). Household income values are adjusted for inflation according to the household’s month in the survey, using twelve-month averages of the CCPI. For instance, income from households reporting fresh income data for October 2023 (i.e., those in months 1 and 5) is adjusted using the November 2022 through October 2023 CCPI twelve-month average. The income from the remaining groups in the sample are adjusted accordingly.
3. Computation of the median household income
The essence of this step is straightforward. Nominal and real (i.e., inflation-adjusted) medians are calculated each month. For each monthly CPS sample, households in the universe are ordered from the lowest to the highest household income (either nominal or real), and a weighted cumulative count of households is assigned to each household record. The household income value that falls in the middle of the weighted cumulative count represents the weighted median household income for that specific month.
Every October, medians are benchmarked based on the official estimates published by the Census in September, adjusting our most recent March nominal value to correspond with the official nominal value reported for the previous calendar year. Real median values are then adjusted to correspond with the benchmarked nominal values.
4. Seasonal adjustment
The series of median household income data are adjusted for seasonality using the X-13ARIMA-SEATS software, which is produced, distributed, and maintained by the Census. The Census X-13 software is widely used for seasonal adjustment and time series analysis of economic and demographic data.
The main purpose of the X-13 software is to remove seasonal and calendar-related variations from time series data, allowing analysts to better understand underlying trends and make more accurate comparisons between different time periods.
© 2023-2024 Motio Research LLC