WASHINGTON (AP) — The 2010 census missed more than 1.5 million minorities after struggling to count black Americans, Hispanics, renters and young men, but was mostly accurate, the government said Tuesday.

The Census Bureau released an extensive assessment of its high-stakes, once-a-decade headcount of the U.S. population. Based on a sample survey, the government analysis has been a source of political controversy in the past over whether to "statistically adjust" census results to correct for undercounts, which usually involve minorities who tend to vote Democratic.

The findings show the 2010 census over-counted the total U.S. population by 36,000 people, or 0.01 percent, due mostly to duplicate counts of affluent whites owning multiple homes. That is an improvement from a census over-count of 0.5 percent in 2000.

However, the census missed about 2.1 percent of black Americans and 1.5 percent of Hispanics, together accounting for some 1.5 million people. The percentages are statistically comparable to 2000, despite an aggressive advertising and minority outreach effort in 2010 that pushed total census costs to an unprecedented $15 billion.

Also undercounted were about 5 percent of American Indians living on reservations and nearly 2 percent of minorities who marked themselves as "some other race."

"While the overall coverage of the census was exemplary, the traditional hard-to-count groups, like renters, were counted less well," Census Bureau director Robert Groves said. "Because ethnic and racial minorities disproportionately live in hard-to-count circumstances, they too were undercounted relative to the majority population."

"Our belief is that without our outreach, our numbers would have been much, much worse," he added.

The South, led by the District of Columbia, Texas, Georgia, Virginia, North Carolina and Florida, was more likely to have people who were missed. The Midwest and Northeast as a whole posted small over-counts.

The findings come after more than 100 cities including New York challenged the official 2010 results as too low.

The Census Bureau, which recently rejected New York's request to revise the city's count, says the latest analysis will not affect the government's official U.S. population tally of 308.7 million but it will be used to improve the 2020 count. Nor will the analysis affect how the federal government distributes more than $400 billion to states for roads, schools and social programs.

"We remain deeply troubled by the persistent and disproportionate undercount of our most vulnerable citizens — people of color, very young children and low-income Americans," said Marc Morial, president of the National Urban League and chairman of the Census Bureau's 2010 Census Advisory Committee. "At a minimum, the census should have the ability to make an adjustment in the official count to ensure that these individuals enjoy the political representation and fiscal resources to which they are entitled."

The Supreme Court ruled in 1999 that federal law barred the use of sample surveys to adjust census results for purposes of allocating House seats; it left the door open to adjustments for other uses such as congressional redistricting or distribution of federal funds. Shortly after taking office in mid-2009, Groves ruled out statistical adjustments in 2010 for redistricting, citing a lack of preparation time.

On Tuesday, the Census Bureau noted how its efforts to count U.S. residents have improved over time. An undercount of the total U.S. population reached as high as 5.4 percent in 1940, the first time the accuracy of a census was formally measured, and then gradually decreased before an over-count was posted in 2000. American blacks are still the most likely to be missed; their undercounts have improved from a high of 8.4 percent in 1940 but at a slower pace than that of whites.

The government takes a census survey every 10 years. The bureau sends census takers and questionnaires to every U.S. household, though not everyone responds. In 2010, the government faced special challenges of counting transient families displaced by widespread mortgage foreclosures, non-English speaking immigrants fearful of enforcement raids and distrustful Americans opposed to government surveys.

Among the findings:

—Renters were undercounted by 1.1 percent, while homeowners were over-counted by 0.6 percent.

—Broken down by age, men 18 to 29 and 30 to 49 were more likely to be missed in 2010 than other age groups, while women 30 to 49 were over-counted; that is a pattern consistent with 2000. Adults 50 and older had over-counts of their population, while some young children ages 4 and under were missed.

—The District of Columbia had the highest shares of people who were missed, at 2.2 percent. West Virginia had the highest over-count of its population, at 1.4 percent.

Democrats and Republicans for years have disagreed on whether the census should be based on a strict head count or cross-checked against a "statistical adjustment" to include hard-to-track people, particularly minorities, who might have been missed.

While statistical adjustment was ruled out in 2010, the issue could re-emerge for the next census in 2020 as the agency struggles to find ways to rein in ballooning costs to its traditional paper mailing operation while striving to reduce a persistent undercount among minorities. Groves is leaving his post in August to become provost at Georgetown University, and in an election year President Barack Obama has not yet nominated his successor.

"Given the difficult economic climate leading up to the 2010 census — with Latinos hit particularly hard by foreclosures, job-loss and other issues that can hinder census participation — the results of this census could easily have been far worse," the National Association of Latino Elected and Appointed Officials said in a statement. "As the U.S. population becomes more diverse, it is increasingly important for the country to invest in funding that will allow the Census Bureau to address the undercount challenges."