The US economy added a stronger-than-expected 147,000 jobs in June, and the unemployment rate ticked down to 4.1%, according to this morning’s Bureau of Labor Statistics report.
As has become depressingly predictable, this release was immediately followed by political noise from both sides of the aisle.
The Trump administration has been proposing rules that would make it easier to fire federal workers, including BLS economists, raising concerns about potential political pressure on data reporting.
Meanwhile, Democrats continue to point to Trump’s baseless claims from August that the Biden administration was “fraudulently manipulating job statistics” as evidence of his willingness to undermine trust in official statistics.
But here’s the thing: I don’t care about the politics. I care about the data. And more importantly, I care about what an independently estimated econometric model tells us about the reliability of that data.
Let me walk you through the model I’ve estimated.
This is a straightforward OLS model of monthly changes in US private-sector employment (USPRIV). I focus on the private sector because it’s less susceptible to policy-driven discontinuities than total employment, which includes government hiring.
The model specification is:
Δ₁m USPRIVₜ = α + Σᵢ₌₁¹² βᵢ Δ₁m USPRIVₜ₋ᵢ + γ₀ Δ₁m ADPₜ + γ₂ Δ₁m ADPₜ₋₂ + εₜ
Where:
- The dependent variable is the monthly change in private employment (in thousands).
- The model includes 12 lags of the dependent variable to capture persistence and momentum.
- It incorporates changes in ADP private employment: contemporaneous (
t) and two months lagged (t-2), reflecting both immediate and delayed payroll effects. - The model is estimated on data from January 2010 through December 2019 and January 2022 through December 2024.
- All forecasts from January 2025 onwards are fully out-of-sample.
The Results Are Statistically Stable and Economically Intuitive
The model performs well:
- Root MSE is approximately 80,000 jobs, indicating tight forecast precision.
- Key predictors include:
- ADP lag 2 with a stronger coefficient than contemporaneous ADP, highlighting delayed effects in payroll data.
- A positive constant, consistent with the underlying trend in private-sector job growth.
- The 12 lags of
USPRIVcapture the typical momentum and reversal patterns in monthly employment dynamics.
The model tracks the evolution of employment remarkably well and closely matches actual BLS data through both the Biden and Trump administrations.
What About Forecast Accuracy Under Different Administrations?
To assess whether employment figures under either the Biden or Trump administrations exhibit signs of manipulation or structural breaks, I’ve analysed forecast errors across three distinct periods: the pre-COVID decade (2010–2019), the post-COVID Biden term (2022 to January 2025), and the early months of Trump’s second term (February to June 2025).
| Period | Avg. Error (k) | Std. Dev. | N Obs | Std. Error | Error/Std |
|---|---|---|---|---|---|
| Pre-COVID (2010–2019) | +9.8 | 71.4 | 107 | 6.9 | +0.14 |
| Biden (2022–2025/01) | –1.4 | 124.8 | 37 | 20.5 | –0.01 |
| Trump (2025/02–06) | –38.6 | 75.8 | 5 | 33.9 | –0.51 |
The model tracks actual employment changes with impressive consistency.
Under Biden, the average forecast error is virtually zero. Under Trump, the model slightly underpredicts job growth, but the deviation remains modest and statistically insignificant.
Crucially, the errors observed during both administrations fall well within the range seen during the stable, pre-pandemic decade.
There is no evidence of systematic bias and no indication of data manipulation. The BLS figures continue to behave exactly as one would expect based on long-standing empirical relationships.
The June 2025 Divergence: ADP vs BLS
This morning’s employment report offers a fascinating snapshot of how different data series can tell distinct stories—without suggesting any manipulation.
ADP reported that private-sector payrolls fell by 33,000 in June—the first decline since March 2023. In contrast, the BLS establishment survey recorded a gain of 74,000 private-sector jobs.
My model, which combines ADP data with lagged private-sector employment dynamics, predicted an increase of 71.6k. With the actual coming in at 74k, the forecast error was just 2.4k—well within the model’s expected range.
The gap between ADP and BLS isn’t a sign of data inconsistency or foul play.
It reflects different methodologies: ADP is based on a single payroll processor’s clients, while BLS surveys a broader swathe of employers and applies different seasonal adjustments.
But here’s the twist: the real surprise in June wasn’t the private sector at all—it was the public sector, which added 73,000 jobs. That surge came mainly from state and local government hiring in education and healthcare—categories that ADP doesn’t cover at all.
So despite political rhetoric about shrinking government or civil service freezes, public employment continues to grow—even under Trump. If the plan was to slow down hiring via executive order or DOGE-themed disruption, it hasn’t shown up in the numbers yet.
By blending the signal from ADP with historical labour market persistence, the model offers a more stable and comprehensive benchmark.
And once again, the results are clear: no anomalies, no bias—just a labour market evolving along predictable, statistical lines.
Visual Evidence: The Model Tracks Reality
The chart above tells the story better than any statistical test.
There’s no structural break, no sudden deviation, no evidence of systematic bias under either administration. The data continues to evolve exactly as the model would predict based on historical relationships.
Trust the Process, Not the Politics
She’s right to be concerned about institutional safeguards. But the empirical evidence shows that, so far at least, the data integrity remains intact.
The accusations flying from both political camps – whether Trump’s claims of Biden “fraudulently manipulating job statistics” or current concerns about Trump undermining the BLS – simply don’t survive contact with the data.
My model provides a neutral benchmark. It doesn’t know or care who’s in the White House. It simply captures the statistical relationships between different employment measures over time. And it continues to track the official data with remarkable accuracy.
The labour market might actually be softening a bit
And the markets don’t seem particularly rattled by the data either.
Yes, June’s jobs number came in well above the ADP estimate, but the market reaction has been muted. Bond yields are up slightly, but in the context of this year’s volatility, a 5bp move in 10-year Treasuries barely qualifies as news.
What deserves more attention, however, is the trend captured by the model—based on both BLS and ADP data. This combined signal offers a more stable and perhaps more reliable indicator of the true trajectory of the US labour market than either source alone.
And here’s what stands out: while the model was broadly flat to up through the second half of 2024, it has clearly begun to trend downward since February 2025.
That shift points to a moderate but persistent softening in private-sector job growth—entirely consistent with a labour market that remains resilient, but is gradually cooling.
So while the Trump administration may be celebrating today’s headline beat, and Fed hawks might use the data to argue against further rate cuts, the underlying message is more nuanced.
According to the data that actually tracks labour market fundamentals, momentum is easing—not accelerating.
Conclusion: Economics, Not Politics
Good policy requires good data. But good data requires trust – trust that the numbers reflect economic reality, not political convenience.
The evidence from my model is unambiguous: that trust remains justified. Neither the Biden administration nor the Trump administration has successfully “cooked the books.”
The Bureau of Labor Statistics continues to produce reliable, professional statistics that pass the test of independent validation.
