The survey focuses on how voters view the economy and also explores attitudes on two hot-button issues over the summer: immigration and race relations.

POLITICO again partnered with SocialSphere Inc. of Cambridge, Massachusetts, to survey likely voters in states with competitive Senate elections and in competitive House districts. The poll is designed to track the opinions of likely voters in those states and districts over the final six months of the 2014 campaign, in addition to exploring individual policy areas with greater depth in a particular survey.

SocialSphere Inc. designed and directed the survey, interviewing 917 likely voters in the states and districts with the most competitive Senate and House races, as ranked by the University of Virginia Center for Politics. Conducted Aug. 29-Sept. 7, the survey was administered on the web by GfK, using a sample from KnowledgePanel, a large, national probability-based online-survey panel. Respondents completed the online survey in English and Spanish; the average completion time was eight minutes.

( Also on POLITICO: Sanford, fiancée break engagement)

While the polls are completed online, KnowledgePanel’s sampling frame includes Americans living in households without Internet access. GfK sends mailers — in English and Spanish — to random addresses in the U.S. Postal Service’s database. Recruits can join the panel by mailing back a form, calling a toll-free phone number or visiting a website. GfK provides those who want to join the panel but don’t have Internet access with a computer and/or free Internet access.

This sampling method addresses some criticisms of online polls — namely, that respondents are self-selecting. Unlike so-called opt-in panels — for which firms recruit members via pop-up ads or ads on websites and which anyone can join — KnowledgePanel is comprised only of those Americans randomly selected through GfK’s address-based sampling. A design summary for KnowledgePanel is available here.

The margin of error for the full sample is plus or minus 3.7 percentage points. For likely voters in competitive Senate states, the margin of error is plus or minus 5 percentage points. For likely voters in competitive House districts, the margin of error is plus or minus 5.3 percentage points.

In order to account for movement on the battleground map, there were some small changes in the sampling frame from the first survey.

( Also on POLITICO: Dems push for immigration delay)

With Kansas Sen. Pat Roberts looking vulnerable after a bruising primary, the Sunflower State was added to the list of Senate states surveyed in this wave. Meanwhile, three House districts in which Republicans look like safer bets were taken off the list of House races: Michigan-11, West Virginia-01 and Utah-04. And with well-funded Democrat Kathleen Rice increasing her grip on New York-04, that seat came off the board as well.

With these modest changes to the sampling frame, comparisons with the first two surveys — the generic-ballot question has moved from an 8-point Republican advantage in our first wave to a 1-point Democratic edge now — are imperfect. Changes to the universe of potential voters could account for some of the movement from poll to poll.

Likely voters were surveyed in the following states with competitive Senate elections: Alaska, Arkansas, Colorado, Georgia, Iowa, Kansas, Kentucky, Louisiana, Michigan, Minnesota, Montana, North Carolina, New Hampshire, Oregon, South Dakota, Virginia and West Virginia.

Likely voters were surveyed in the following competitive House districts: Arizona-01, Arizona-02, Arizona-09, Arkansas-02, Arkansas-04, California-07, California-10, California-21, California-26, California-31, California-36, California-52, Colorado-06, Connecticut-05, Florida-02, Florida-18, Florida-26, Georgia-12, Illinois-10, Illinois-11, Illinois-12, Illinois-13, Illinois-17, Indiana-02, Iowa-01, Iowa-02, Iowa-03, Iowa-04, Maine-02, Massachusetts-06, Michigan-01, Michigan-07, Michigan-08, Minnesota-02, Minnesota-07, Minnesota-08, Montana-AL, Nebraska-02, Nevada-03, Nevada-04, New Hampshire-01, New Hampshire-02, New Jersey-02, New Jersey-03, New Mexico-02, New York-01, New York-11, New York-18, New York-19, New York-21, New York-23, New York-24, Ohio-06, Ohio-14, Pennsylvania-06, Pennsylvania-08, Texas-23, Virginia-02, Virginia-10, West Virginia-02, West Virginia-03 and Wisconsin-06.