You interviewed three people, liked one, hired her, and she was gone by spring. Every director has made that call and been wrong about it.
The good news is that hiring accuracy is one of the most researched subjects in the workplace, and the findings are cheap to apply. The awkward news is that the part most centers focus on, which questions to ask, matters least.
So here are twenty questions worth asking. First, the two things that decide whether they tell you anything.
Same questions, scored answers
In 2022 a team led by Paul Sackett re-ran the numbers behind decades of hiring research and found the old estimates had been inflated. Structured interviews came out top of every method they could re-score, at .42. Unstructured interviews came in at .19, and the bottom of that range dips below zero, meaning in some workplaces they predict nothing at all.
A structured interview isn't complicated. Every candidate gets the same questions in the same order, and you score each answer against a scale you wrote beforehand.
An older study sorted interviews by how much structure they had. Loose questions plus a general impression sat at the bottom. Same questions plus scored answers sat near the top, and tightening further, banning follow-ups altogether, added almost nothing. So you can still ask someone to say more. You just can't wing the whole thing.
What centers screen on instead
Two economists sent 10,986 fake resumes to 2,772 childcare job ads across fourteen American cities and recorded who got called back.
Applicants with a bachelor's in early childhood education were no more likely to get an interview than applicants with an associate's.
Experience helped, but only up to a point. Six months in early childhood raised the callback rate by 14 percentage points against no experience at all. Two years raised it by 11.5. The applicant with two years was significantly less likely to be called than the one with six months.
Which is worth sitting with, because years of prior experience predicts job performance at about .07. Close enough to nothing that you shouldn't sort resumes by it.
A companion paper from the same experiment found something harder to read. Resumes with Black-sounding names drew interviews 20.1% of the time, against 27% for white-sounding names. In states with stricter staffing rules and quality ratings that gap was much smaller. That's a correlation rather than proof, but structure keeps pointing the same way.
The twenty questions
Ask all twenty, in this order, of everyone. Most are past-behavior questions, which predict better than hypotheticals here.
Safety and judgment
1.Tell me about a time you noticed something unsafe in a room. What did you do?
2.You're alone with eight two-year-olds and one starts choking. Walk me through your first thirty seconds.
3.Tell me about something you had to report that you'd rather not have reported.
4.Describe a time a child got hurt on your watch. What did you tell the parent?
5.What would make you fetch a director rather than handle it yourself?
The work itself
6.Describe your last full morning with a group of children, hour by hour.
7.Tell me about a child who was hard to settle. What did you try, and what worked?
8.What do you do with a child who bites?
9.Tell me about an activity you planned that fell flat.
10.How do you handle nap time when three children won't sleep?
Parents and colleagues
11.Tell me about a difficult conversation you had with a parent.
12.Describe a time you disagreed with a co-teacher about how to handle a child.
13.What's the hardest thing to say to a parent at pickup?
14.Tell me about a time you covered for someone at short notice.
15.Describe feedback from a director that you didn't agree with.
Whether they will stay
16.What made you leave your last role?
17.What does a good day here look like to you?
18.Which age group would you pick, and why that one?
19.What do you expect to find hardest in your first three months?
20.What would we need to do for you to still be here in two years?
Then add up the scores
Score every answer as you go, one to five, against notes you wrote in advance. Then total them.
That last step is the one centers skip, and it's expensive. A meta-analysis comparing the two ways of reaching a decision found that adding scores up predicted job performance at .44, while talking it through and forming an overall view came out at .28. That's better than a third of your accuracy, thrown away in the debrief, even when the people talking know the job inside out.
So run the conversation, then let the numbers decide. If the highest total isn't the person you liked, that's the point of having done it.
Two more things. If you run a trial shift, and you should, pay for it. The Labor Department has made employers pay back wages for unpaid working interviews. And a new hire can start once either the FBI fingerprint check or your state's criminal registry check clears. Until the rest come back she must be supervised at all times. States get up to 45 days.
Treat reference checks the same way. There's no credible modern estimate of how well they predict performance. Run them to catch what you must catch, not to find out who'll be good with toddlers.
The part after the offer
Getting the hire right only pays off if you keep track of what happens next. Those checks have to be repeated at least every five years, and every credential has a date on it.
ChildcareWow tracks licenses and credentials with the expiry dates attached, and keeps staff records and schedules in one place. It started there, because a licensing alert that never got sent is what made us build this.
It won't interview anyone for you, and it won't tell you who to hire. It will stop a clearance quietly lapsing on someone who's been with you four years.
We built it because we run centers too.
Book a 20-minute walkthrough(https://booking.childcarewow.com/#/Demo). We'll show you your staff records and expiry dates the way your director would see them.
SOURCES
• Sackett, Zhang, Berry & Lievens, "Revisiting meta-analytic estimates of validity in personnel selection," Journal of Applied Psychology 107(11), 2022 — structured interviews .42 (top-ranked of all procedures re-scored), unstructured .19 with 80% lower credibility value of -.01, work samples .33, cognitive ability .31, years of experience .07 — https://doi.org/10.1037/apl0000994
• Kuncel, Klieger, Connelly & Ones, "Mechanical versus clinical data combination in selection and admissions decisions: A meta-analysis," Journal of Applied Psychology 98(6), 2013 — job performance criterion, k=9, N=1,392 mechanical / 1,156 clinical. Mechanical .44 vs holistic .28, "even by experts who are knowledgeable about the jobs and organizations in question" — https://doi.org/10.1037/a0034156
• Boyd-Swan & Herbst, "The demand for teacher characteristics in the market for child care: Evidence from a field experiment," Journal of Public Economics 159, 2018 — 10,986 resumes to 2,772 childcare job ads across 14 US cities, May 2016 to Jan 2017. BA in ECE no better than an associate's; six months of experience +14.0pp vs none, two years +11.5pp, a significant 2.5pp penalty for two years against six months — https://doi.org/10.1016/j.jpubeco.2018.02.006
• Boyd-Swan & Herbst, Educational Researcher 48(7), 2019 — SAME experiment as above. White-sounding names 27% interview rate, Black-sounding names 20.1%, Hispanic 23.8%. State childcare regulations and quality certification "mitigate or eliminate entirely" the gap; a one-SD rise in the regulatory index raises Black applicants' interview likelihood 4.3pp. CORRELATIONAL, not experimental — https://journals.sagepub.com/doi/10.3102/0013189X19867941
• Van Iddekinge, Arnold, Frieder & Roth, "A meta-analysis of the criterion-related validity of prehire work experience," Personnel Psychology 72(4), 2019 — k=32, N=8,463, corrected r = .07 for PRE-HIRE experience — https://doi.org/10.1111/peps.12335
• Huffcutt & Arthur, "Hunter and Hunter (1984) revisited: Interview validity for entry-level jobs," Journal of Applied Psychology 79(2), 1994 — structure gradient: Level I .20, Level II .35, Level III .56, Level IV .57. Level III to IV adds only .01, which is why the post says you can keep asking follow-ups — https://doi.org/10.1037/0021-9010.79.2.184
• 45 CFR 98.43 (CCDBG), current text including the 1 March 2024 final rule. §98.43(d)(4): a prospective staff member may begin work after a qualifying result on EITHER the FBI fingerprint check (b)(1) OR the state criminal registry check in their state of residence (b)(3)(i), and "must be supervised at all times" until all components clear. §98.43(e)(1): states must complete a check "not to exceed 45 days" — https://www.ecfr.gov/current/title-45/subtitle-A/subchapter-A/part-98/subpart-E/section-98.43
• U.S. DOL Wage and Hour Division news release 18-1934-ATL, 14 December 2018 — Smiley Dental Associates, Nashville, required candidates to perform an unpaid "working interview"; $50,000 in back wages and liquidated damages for 10 employees, alongside other wage violations — https://www.dol.gov/newsroom/releases/whd/whd20181214
