The Index draws on unique proprietary data from Care.com(1) as well as publicly available data from other organizations including the U.S. Census Bureau, U.S. Bureau of Labor Statistics, the National Association for the Education of Young Children (NAEYC), and the National Association for Family Child Care (NAFCC) from 2015. (In order to provide a consistent snapshot of paid child care arrangements across the course of the year (school-year and non-school year), all data are taken from the full 2015 calendar year.) A new Care.com survey, conducted in October, 2015 by Hanover Research (a market research company), using an online panel maintained by ResearchNow (a digital data collection company), with design input from New America, provides additional data on paid child care arrangements in each state with a sample of households with children.(2) Survey questions include detailed information about the primary childcare arrangement for each child, time to find a caregiver, and reasons for changing child care providers. We use the index as exploratory research to identify key child care challenges and take a deeper look into what's happening on the ground.
(1) The Care.com dataset is an important new asset and lens into the overall care market; Care.com currently hosts 18.4 million members of which 56% are families seeking care and 44% are caregivers. Care.com users are present in all states and distributed roughly equal to the overall population distribution by state. In 2015, Care.com hosted nearly 7 million caregivers, families in approximately 93% of U.S. zip codes and caregivers in approximately 88% of U.S. zip codes. The typical care seeker for U.S. consumer business is female (82%), has an average household income of $75,000, and has at least one child under 18 in the house (73%). The typical caregiver is also female (94%) and well educated (61% indicating they have at least some college education). Care.com members seeking child care are similar to the U.S. population of families with children when it comes to the percentage of women who are the primary child care decision-makers and the number of families with dual incomes.
Nonetheless, Care.com members skew older than first-time moms, are somewhat more likely to live in an urban area, and have higher incomes on average. The higher income-skew is likely due to the fact that fewer low income families pay out-of-pocket for full-time child care. The Care.com survey was issued to more than 1,000 respondents in the "Less than $25K" income category in order to compare how low income households differ from high income households.
(2) Sampling frame, technique and execution: The Care.com survey uses household contacts available from online panel company ResearchNow, whose panel consists of 2 million people. The survey targeted a sample of online panel members in households with children under the age of 18. Quotas were set to ensure 385 valid responses per state (with additional quotas by income, race and marriage), and a margin of error of 5% at a 95% level of confidence. More specifically, the survey was sent out in three waves:
Wave one: households with children under 18 with quotas by race, income, and state.
Wave two: married couples with children under 18 with quotas by race, income, and state.
Wave three: persons age 18-64 with children under 18 (no quotas).
The response rate was 6.25%, meaning the percentage of opened survey invitations was 6.25%. It is possible that someone was surveyed more than once; ResearchNow was only able to ensure that each email address was only surveyed once. Total number of survey respondents was 15,038.
To prepare the data for state-level analysis, American Community Survey (ACS) Public Use Microdata Sample (PUMS) data were used to match the distribution of responding households on key characteristics to the state population distributions on those measures. If sample distributions on key household characteristics did not match state-level distributions of household characteristics, the data was weighted to eliminate known biases. The survey was weighted on income, marital status, and employment status.
To prepare the data for national level analysis, further weights were added based on 2010 U.S. Census data detailing the number of households with children under 18 per state. In other words, responses from states with higher proportions of total U.S. households with children under 18 were given greater weight when conducting national-level estimation, compared to responses from states with lower proportions of total U.S. households with children under 18.
This approach makes the survey responses representative of state and national distributions for these variables. Nevertheless, it does not account for any bias that could arise from non-random sampling. The known limitations of the survey are as follows:
Weighting data at the state and national level will approximate a representative sample. However, utilizing a voluntary panel rather than a probability-based panel, or random sampling through random digit dialing, introduces bias.
Selection bias on unobserved variables is not minimized.
Random sampling error cannot be estimated.