Never mind how you define quality.

Are there certain things stations can do that will attract viewers?

Last year we tried to outline such a "magic formula." We used a method common in business – identifying models of success. Using four years of data, we found stations that were building viewership and isolated what distinguished them from stations failing to win viewers. It was basic "benchmarking."

This year we are adding a second approach. Rather than just comparing one group of stations to another, we also used regression analysis, a statistical method that tries to predict outcomes, to isolate what works.

At a certain commonly accepted level of statistical significance, this method can identify practices that are particularly effective in attracting and holding viewers. We measured these approaches against four different criteria of economic success – ratings, share, retention of lead-in audience and demographics. (The share and ratings data are for five years. The analyses of lead-in audience and demographics are just for 2001 and 2002.)

Using this more rigorous statistical approach, some of our findings from last year were strengthened, others were not, and additional findings emerged.

Do More Enterprise Reporting

Effort pays.

Over five years, stations with better ratings and shares do more special series. Stations with better audience retention and key demographics do more tough interviews.

When the more involved statistical analysis is applied, the case for enterprise becomes even stronger.

Stations that do more enterprise of all sorts are more likely to build on their lead-in audience. Enterprise alone added 6 percent to the rate at which a station held its lead-in audience.

Conversely, stations that produced the fewest enterprise stories (the lowest 20 percent of stations) had worse than average lead-in retention (2.5 on a 5 point lead-in scale), while those in the highest 20 percent scored significantly above average lead-in retention at 3.15.

We measured the enterprise level of every story – from original investigations at the top, to sending a reporter and not just a camera in the middle, all the way down to using video press releases.

The statistical model shows that all kinds of enterprise helps retain audience. Doing substantive on-camera interviews alone can add 4 percent to the rate at which a station holds its lead-in audience. To retain viewers, do more original stories, not just the daybook. Don't air so much feed tape. Connect events to your community. Send a reporter, not just a camera, and your audience will stay engaged.

Unfortunately local news is moving in the opposite direction. Even sending a reporter out to cover a breaking story or an event from the daybook has become the exception rather than the rule: in 1998 these two categories accounted for 62 percent of all stories; now they account for just 37 percent. The most common story format now is the anchor voiceover, averaging 29 seconds. Four years ago anchor voiceovers made up 9 percent of all stories; that figure has now leaped to 29 percent.

There are other signs that enterprise is declining. Satellite downloads have risen to 23 percent of all stories, up from 14 percent in 1998. Original investigations, tough interviews, and even reports on other people's investigations have nearly disappeared, especially on late-night newscasts.

Is there a silver lining? Yes. When stations did stories with original reporting, over the past four years their stories scored increasingly higher for relevance to the local audience. Stations know how to do work that viewers care about when they have the time and resources to do it.

Cover more of the community

Newscasts that air more locally relevant stories are significantly more likely to hold onto or attract a larger audience than the preceding program. Statistically, producing important local stories can add 1 percent to the audience lead-in retention rate.

The statistical correlation here was strong. It didn't apply to ratings and share, but making stories locally relevant clearly helps keep lead-in audience.

Does this mean local stations should do only local news? Hardly. But it means that there is clearly a right and wrong way to do national and regional news that will build audience.

While it may surprise some broadcasters, the right way is to "localize" national issues by finding community examples, getting local comment and clarifying the issue's local impact: How does President Bush's "Leave No Child Behind" education bill affect the community?

The wrong way is to find stories that aren't necessarily significant from a civics standpoint but may contribute to the "water cooler" discussion at the office the next day – the Robert Blake murder case, say, or the San Francisco dog mauling trial. These stories, the thinking goes, are so fascinating that they require no localizing.

The data clearly show that the "water cooler" approach – doing more stories that have buzz factor – doesn't work.

Unfortunately, that approach dominates. Stations today are three times more likely to report national stories without a local context (13 percent of all stories) than they are to report national stories with an explanation of the local consequences (4 percent). The Blake case and the dog mauling trial were among the biggest "buzz" stories in the 2002 study. That type of reporting loses viewers.

Some news directors apparently have reached the same conclusion. One news director told our survey that more "community impact" is his strategy for improving ratings. Placing "emphasis on issues affecting the community," wrote another. "It may sound cliché but our previous management got away from that and our ratings took a turn for the worse."

Last year, using benchmarking, we also found that stations with better ratings, share, demographic, and lead-in trends were more likely to air a wider range of topics in their newscasts. That still holds true with the fifth year of data. However, the pattern was not reinforced by the regression analysis. We cannot show statistically that adding more topics to a newscast will itself help add viewership.

But the statistical analysis did find something else. Stations that air the most highly substantive stories – those that focus on ideas, issues, or policy, or investigate public malfeasance – are significantly more likely to retain their lead-in audience.

This doesn't mean you should just air more political stories. Rather, it relates to how you cover your community. Look for the long-lasting impact and deeper issues in stories. Don't be afraid to investigate the powers that be. The audience pays attention.

Air More Long Stories

Every year of the study, we've found that successful stations were more likely to air longer stories.

On average, for example, the stories on stations with better demographic and lead-in trends are five seconds longer.

Now the case for longer stories is further confirmed by the more rigorous statistical analysis.

Even though fast pacing is assumed to attract viewers, stations that air more long stories are likely to improve their demographic trends by 4 percent.

Put another way, viewers like stories they can sink their teeth into.

Why? Longer stories tend to contain more points of view, involve longer soundbites, show viewers more pictures, go more places, get into more detail, and give viewers more time to comprehend their points.

Rather than throwing facts at people, long stories can tell a tale.

Source Stories Better

Viewers notice sources.

Last year we found that stations with successful viewership trends were less likely to air stories citing anonymous sources or no sources at all. They were also more likely to air stories with multiple sources and to feature sources with higher levels of expertise.

With the fifth year of data added, this finding is still true, though the margin of difference shrank slightly.

In addition, this year's more refined statistical analysis shows that viewers particularly like hearing from people who know what they're talking about. Stations that relied on more authoritative sources in stories improved their hold on their lead-in audience by 10 percent.

Similarly, stations that avoid unattributed or unsourced stories improve audience retention and build key demographics.

We measured the authority of sources by noting their level of expertise for a given story – from an appropriate "expert," to a participant, to citing no source at all. Thus a doctor with the right specialty is an expert in a medical story. A commuter may be the expert for a story on traffic congestion.

There is plenty of room for improvement here, as well as reason for optimism.

In the first four years of our study, only 16 percent of the stories included comments from an appropriate "expert."

This year, however, the use of experts rose to 22 percent. Some of this increase can be attributed to the number of stories about the war in Afghanistan, a campaign reported largely through on-camera briefings by Secretary of Defense Donald Rumsfeld and others with military expertise.

The challenge is finding ways to get more authoritative sources into the kind of stories local news does more often – not just Pentagon press briefings.

Hire More Reporters and Give them More Time

Stations that invest their money in more people do better.

This finding has been reinforced in several ways. Three years ago we discovered that stations putting more of their budget toward staff, and less toward equipment, had better ratings trends. Last year, we made a related discovery, that stations that give people more time to work on stories have better ratings trends.

This year we again found evidence that giving the reporter more time helps – in both ratings and share. The most successful stations assigned an average of 1.5 or fewer stories to reporters each day, while the average overall was 1.8.

We have another new finding related to budgeting. Stations that had not asked reporters to do multiple versions of their stories this year and were spending more on production staff were also having more commercial success. On average, they were one full point higher on the five-point scales used to measure rating and share.

Tom Rosenstiel is director of the Project for Excellence in Journalism. Marion Just is a professor of political science at Wellesley College and a research associate at the Shorenstein Center on the Press, Politics, and Public Policy at Harvard.