8
Grading Your Programming Performance: What You Need to Know about Ratings
Once upon a time, radio ratings simply estimated the number of listeners age twelve and over (“12+”). In fact, there was a time when they were commonly presented in terms of all listeners age six and over. General circulation newspapers still like to quote radio’s audiences in terms of “12+ shares,” and you’ll usually find them summarized that way in trade magazines, too. Today, though, we are much more concerned with demographics, or specific age groups, and even “psychographics,” which refers to lifestyle and attitude.
The reason for this is simple. Ratings are primarily a sales tool, and these days advertisers want to target their advertising to specific groups of people. In fact, the ability to target narrow age groups and lifestyles is one of radio’s strengths. Therefore, the ratings for “everybody age 12 and over” are essentially irrelevant today to any radio station’s goals, although it’s always nice to look good in a ratings display, even that one.
Unfortunately, as radio stations—and advertisers—have become more and more focused in targeting smaller and smaller segments of the population, the ratings become more and more inaccurate. It’s not that the research companies are getting sloppier; it’s that we are placing more and more burden on their numbers—and the more narrowly we break down individual segments of the ratings, the less confidence we can have on the numbers that result.
Here’s why. Radio ratings have been based, since the twenties, on the same technique used in political and opinion polling. Newspapers and magazines are able to develop circulation figures based on what is sent out and what is returned unsold, subtracting the latter from the former and coming up with a hard figure. Of course, this doesn’t tell how many of the copies are actually read, which sections are read the most, how many readers there are per copy, and how many of the readers actually read the ads. To obtain this information, the printed media have to resort to the same sort of methodology that radio and television use. Newspapers, in particular, usually want to avoid getting into such unsettling subjects at all because they would be forced to admit that some parts of the paper are read more than other parts. The newspaper ad pricing structure is still based on maintaining the fiction that advertisers achieve similar benefits regardless of where their ads are placed.
Radio and television, however, leave no trail as they depart the transmitter at the speed of light, and there is nothing in the signal to tell how many receivers are tuned to a specific station. Advertisers wanted circulation information, and the opinion-poll technique was pressed into service. The basis of this technique is the law of probability, which in essence states that if you randomly select a small number of people from a given population and determine what they think or do, you can then project the “ratios” (or percentages) of opinion or behavior you find against the whole population within a mathematical percentage of confidence. The larger the sample of the population, the higher the degree of confidence you can have in the results (the smaller the margin of error in the data, in other words).
Of course, if the sample you collect is not a true random cross section of the population, the results become less reliable the less random it is. (However, if it is a random sample of an identifiable subgroup, it can be pretty accurate in representing that subgroup.) We’ll spend some time later in this chapter looking at how this problem can cause real trouble in radio ratings results.
The other problem that has developed from using a polling method that was originally limited to yes or no answers, or to selecting a preferred candidate from a very limited number of choices, is that— as time has gone by—more and more radio stations (choices) are measured in each survey. In even small markets, there may be twenty or more stations listed in the rating book; in large markets, more than fifty stations may be listed. This means that now a very small fraction of that random sample accounts for all the data collected for each individual station, and the smaller the sample used in any polling, the larger the error factor must be.
Now, consider that we have pages and pages of ratings data in each survey to show how each of the stations performs within specific segments of the population, defined by age group and gender. Because ratings companies print data showing how each of many stations ranks in size of audience in such narrow segments of the population as “men, ages 25 to 34, Saturday morning 6 to 10 A.M.,” you can imagine how accurate that sort of data are going to be. In some surveys, you’ll see the same “share” numbers repeating down a column or “share” numbers that are mathematical multiples of others in the same column. This tells you that very, very few respondents were measured in this segment of the audience. The error factor may be in a few hundred percent.
As program directors, what we need to understand about ratings is not only what they can tell us about the audience for our station and others, but also what they can’t. We need to know what the statistical problems with ratings are. How much of what they tell us is real, and how much is “statistical noise”? All too often, stations change staffs (including their program directors) and formats, due to nothing more than a “bad book,” which a little elementary analysis would have shown was very unlikely to be right.
Analyzing Ratings
Because we can’t come to any conclusions about our ratings until we know how much confidence we should have in what they show, this section describes how to do basic ratings analysis.
The earliest form of radio audience research was based on telephone coincidental interviewing. Researchers dialed phone numbers at random and asked those who answered whether their radio was on and, if so, to what station it was tuned. When enough listeners were included in the survey, it was possible to project two figures: (1) the percentage of all households that had the radio on in each quarter hour and (2) the share, or percentage, of total radio listening that each network—later station—received. The second figure later came to be known as the average quarter hour share, and the telephone coincidental method was the only method of directly obtaining it. The resulting number, the percentage of total radio listening that each station receives, is often stated along with the average quarter hour rating, which is a percentage of the total population estimated to be listening to the station in the average quarter hour.
To understand the distinction between these, suppose that only 20 percent of the total population is listening to radio in any average quarter hour. A station with a 10 share of listening, the share figure usually quoted when discussing audience shares, has an average of 10 percent of the total average number of radio listeners in that time period (or “daypart”). However, because only 20 percent of the total population was listening to any radio station in the average quarter hour in this example, the average quarter hour rating, or the percentage of the entire population listening to the station in the average quarter hour, would be only 20 percent of that 10 share, or 2.0. Salespeople usually use ratings points in calculating advertising costs, but these numbers are usually too small to be of any real use in programming.
The C.E. Hooper Rating service used the telephone coincidental method into the sixties before the company disappeared—a victim of its inability to give either demographics or “cume” figures.
Cume—short for cumulative—is equivalent to newspaper or magazine circulation. It counts everyone who hears a station for five minutes or more during a week in the indicated time period, without taking into consideration how long they listened or how often. Although advertising agencies put a lot more weight on the share (because it attempts to report how many listeners might actually have heard a commercial announcement), cume is needed to develop a “turnover ratio,” which can be useful for programmers as well.
The turnover ratio is simply the “cume persons” estimate divided by the “share persons” estimate. It tells how many times the audience turns over, or changes, on average in the time period. This figure can be used to calculate how many commercials must be run for a given client to reach the station’s available audience. A common rule of thumb in the business is that the turnover ratio equals the number of spots that must be run to reach 50 percent of the station’s cume.
This sort of calculation is referred to as “reach and frequency determination”—how many listeners are reached by how many commercials. For example, if a station’s weekly cume is ten thousand people in the selected age group and the average-quarter-hour-persons figure in the same daypart is twenty-five hundred people—and if we’re examining a four-hour daypart, such as 6 to 10 A.M. or 3 to 7 P.M.—the cume persons divided by the share persons would be 10,000 divided by 2,500, which would result in a turnover ratio of 4.
According to the rule of thumb, this means that four commercials, evenly spread throughout the daypart, should reach 50 percent of the cume, or five thousand people. More commercials should reach incrementally more people, but with decreasing efficiency because it would require running the commercial every five minutes to reach all of the cume, some of whom listen for very short periods of time.
This example tells the program director something useful: In this time period, the average listener (in the indicated age group) listens for one hour. We see this by inverting the calculation. Out of ten thousand listeners who tune in the station during the week in this time period, twenty-five hundred are listening in the average quarter hour: 2,500 divided by 10,000 is 0.25, or 25 percent. This means that the average listener is listening for 25 percent of the time in that daypart, which in this case is four hours. Twenty-five percent of four hours is sixty minutes, or one hour.
Unlike C.E. Hooper’s methodology, ratings companies today generally obtain only cumulative information directly, and then they calculate the average quarter hour share by adding up all of the quarter hours listened to by all the listeners who listened to the station at all, and then dividing all those quarter hours by the number of cumulative listeners, to obtain the average listening span. Share is then derived mathematically using this formula:
Average Quarter Hour Persons = Cume Persons × (Average
Listening Span [in Minutes] ÷ Total Minutes in the Daypart)
There are two things you need to understand before you make use of this sort of calculation. First, having had the opportunity of looking through three different survey companies’ underlying data, from which the surveys are made, I can tell you that the average listening span (which, in our example, is sixty minutes) is composed of a handful of people who listen for nearly the whole four hours and a great number of people who listen for only fifteen to thirty minutes per day. So, any plans you make on how often to repeat certain music categories or other features—or how long to make uninterrupted music sweeps, for that matter—must take into consideration the shorter-than-average listening span of the majority of listeners, for maximum impact. That is, if you plan on having your airstaff announce music only every thirty minutes and if your ratings show an average listening span of sixty minutes, you may think that your average listener will hear two announcements. In fact, though, the majority of your listeners will probably hear no more than one announcement and—in a great many cases—none!
This is because of the second point you need to understand: Before ratings companies start to produce either the cume or the share figures, they modify the data by rounding it off in every individual case to the nearest fifteen minutes. Believe it or not, your station gets credit for a quarter hour of listening if a person listens for only five consecutive minutes in that quarter hour. However, if the person listens for four minutes or less, no credit is given at all. In fact, amazingly, if a person listens for five minutes, tunes out, and then tunes back and listens for another five minutes within the same quarter hour, the station will get credit for two quarter hours of listening—in the same fifteen-minute period!
So you see, if your “average listener” listens for sixty minutes per day in that time period, the majority of your listeners may be listening only fifteen to thirty minutes a day. In addition, because a listener is credited with fifteen minutes of listening based on only five consecutive minutes of actual listening, you can be sure that only some of your listeners credited with a quarter hour of listening actually listened the entire fifteen minutes and that a significant number of your listeners will have listened for only five or ten minutes per day.
How, then, can you establish station expectations or identity for people who are part of your cume but who listen for less than fifteen minutes per day? To start, you might want to schedule your breaks more often than you had planned!
However, we haven’t yet decided how accurate your rating “report card” may be, and you really can’t make intelligent programming decisions based on these figures until you understand that. (Unfortunately, for sales purposes, ad agencies usually don’t care about accuracy as long as they have numbers to justify their ad-buying process.)
There have been many requests over the years for larger sample sizes in ratings studies. When so many stations are dividing so small a sample there cannot be sufficient confidence in each individual station’s ratings to make intelligent use of them for programming. This can cause stations’ ratings to wobble up and down from book-to-book, making ratings difficult for sales purposes too. Nonetheless, a larger sample size, though a good idea, won’t solve the biggest problem with today’s radio ratings. There are too many biases and too many departures from the requirements of the law of probability for the samples now used to be a true population cross section.
Understanding the Limitations of Ratings Data
To keep costs under control, ratings companies use sampling methods in which some “sample self-selection” can and does occur. That is, some members of the population are overlooked entirely because the sampling method doesn’t include them, and among groups that are included, participation varies with willingness to participate.
To begin with, any survey that in any way uses telephones for sampling will exclude people who don’t have telephones. Studies show that those who don’t have telephones differ from those who do in various ways—by age, income, ethnic group, and lifestyle. Therefore, excluding nonphone homes will bias the outcome of the survey in favor of some stations and against others. Usually, it seems, ethnic and youth-oriented stations suffer the most from this telephone bias.
Of those people who do have telephones, some keep the number unlisted. If only listed phone numbers are included in the survey, people who are unlisted will be excluded. Studies show that the people who choose to be unlisted are different in various ways than people who choose to be listed. For this reason, all survey companies now try to include unlisted phones in their samples—with varying degrees of success. To whatever extent that unlisted homes are not properly represented in the survey, the data must be further distorted.
Of the people who are contacted by the ratings company, certain types and ages of people are more likely to cooperate than others, and obviously the cooperators are going to be the group represented in the survey. This further distorts the data and further weakens the data’s validity in the very basis of such surveying—the law of probability. In the end, only those who are contacted by the survey company and who choose to cooperate constitute the segment of the population actually randomly sampled—and not only is that a smaller segment of the total population, but it is not a segment that can ever be representative of the population as a whole.
So, should you program just to the people who cooperate in surveys, or should you program to everyone in your demographic target? Ideally, the latter. From a practical point of view, though, the former, but preferably without ignoring the latter.
Currently, the only fully national radio ratings company doing surveys through interviews conducted over the telephone is Willhight Research of Seattle (206-431-8430). They’re the number two ratings service, based on the number of markets served. The company has a special strength in smaller markets due to their lower cost, but it is also attractive to large markets because of the availability of psychographic and product-user data cross-referenced to the ratings.
The number three ratings service is AccuRatings (800-777-8877), which is currently devoting its attention to developing its customer base in major markets. This company uses a controversial telephone interview technique that includes preference questions instead of simply behavioral questions.
The leading rating service is Arbitron (212-887-1300), which uses mail-distributed, self-administered seven-day diaries to record radio listening information. Because Arbitron gathers its data using diaries filled out by the sampled participants, you might think that it sidesteps the problems associated with telephone surveys. Alas, that’s not true. Arbitron uses a telephone-listing sample to place its diaries. In my opinion, this step further compromises their data. Here’s how, based on my years of analysis and study. To begin with, because Arbitron starts with a sample of listed phones, the unlisted numbers are excluded. They obtain a second sample from each market, made by subtracting all listed phones from all possible phone numbers there, and then randomly choose a second sample from those. They then merge the survey data from both samples to compile each market’s ratings report. The problem with this is that they cannot guarantee that they will obtain the same cooperation rate from the listed sample as from the unlisted one. To an extent, then, the two samples are not equivalent, and one can distort the results of the other.
Compounding the problem is that those who are intentionally unlisted tend to have different mind-sets and lifestyles than those who are unintentionally unlisted, so their cooperation levels are quite different, too. The unintentionally unlisted people are those who have moved during the past year; they’re not in the current phone book, but they will be in the next one. Because they don’t think of themselves as unlisted, their cooperation is similar to that of the listed sample. (They are similar to the listed sample, except that they are not as long-established in their address.)
On the other hand, those who are intentionally unlisted—those who pay extra not to be included in the phone book—tend not to cooperate when asked for their name and address by an interviewer who obviously doesn’t know who they are and who has clearly dialed their number by chance. Would you provide this information? They’re paying for privacy. If they won’t tell the caller their name and address, they can’t receive a diary and won’t be included in the survey.
I conclude, then, that Arbitron’s unlisted sample skews toward the unintentionally unlisted and that its participants tend not to be too different psychologically from those in the listed sample. If the intentionally unlisted are indeed underrepresented, this would undercount people who are less gregarious than others and who value their privacy, those who are very rich, and segments of the population that prefer to be unlisted. These people listen to radio, too, but might prefer different stations and listen in different patterns than those who cooperate in the survey.
To those who do agree to cooperate with Arbitron and who do provide a name and address, Arbitron sends a diary. In fact, the company sends a diary for every person in the household age twelve and over, thus clustering reported listening within households. The net effect, in my experience, is an average of about two diaries per household surveyed—and that average might be going up because Arbitron has been trying to improve its response rate from big households, netting more diaries per placement and reducing the cost of diary placement.
The problem with this is that even if every individual member of the household fills out his or her own diary without help—and I’ve seen studies showing that in at least some cases, one person in a family eventually fills out the diaries for everyone in the household—there is nonetheless some shared listening in family situations. I therefore consider that the effective sample size in an Arbitron survey may best be the “households surveyed” number and not the number of diaries in the survey—a substantially smaller number. I feel that, to eliminate this effect, there should be only one respondent per household.
Arbitron asks participants to fill out the diary to reflect actual radio listening for one week. Over the years, some published studies have suggested that the participants who actually record their listening as they listen may constitute as few as half or less of the total participants. In addition, these people may lose interest and log less and less listening as the seven-day week progresses. I find that it’s widely assumed in the industry that the reason that Arbitron always starts surveying on Thursdays is to assure two good weekdays of data, plus enough information to generate weekend listening figures, and that any listening recorded for Monday, Tuesday, and Wednesday constitutes a bonus.
Of course, for those who might fill out the diary only after the week is ended, there is no such tapering off effect, but the respondents might tend to generalize their listening patterns—this is “recall” data—and list mainly the favorite stations they easily remember. This could benefit trendy and well-advertised stations, which are the most easily remembered, and it certainly could raise a cautionary flag about changing a station’s call letters when a format is changed. Often, you are better off keeping the call letters that people have learned and just promoting a format change.
In my experience, there are a couple of other important problems to overcome with Arbitron data. The first is the Hawthorne Effect: When survey participants know in advance that they will be surveyed (such as when they receive a diary in advance), this foreknowledge could change their behavior. Because Arbitron is the only radio survey company that has ever given this sort of foreknowledge to their sample well in advance of surveying, it is uniquely vulnerable to this problem.
My second concern deals with the matter of the “cooperation” of the designated sample. At present, all survey companies have problems with cooperation because all start with telephones in some way or another, and as telemarketing has risen, cooperation has steadily dropped. However, because Arbitron is the only survey company that must learn the name and address of those they call (in order to mail them the diaries), Arbitron is uniquely vulnerable to that particular privacy issue. Also, because other surveys gain the raw listening data they need entirely through the initial telephone interview, it appears that only Arbitron has the additional problem of obtaining further cooperation from those who initially cooperated by accepting diaries but may not cooperate sufficiently to return them. It’s a matter of record in every Arbitron report that only a fraction of those receiving diaries actually return them.
There are a number of other statistical problems with Arbitron data. Because the Federal Trade Commission regulates ratings companies not by requiring accuracy or even compliance with the law of probability, but simply by mandating that any flaws must be listed somewhere in their ratings books, you’ll find in the back of each Arbitron report a long list of “limitations.” I won’t go into those here, but you should read that list and be aware of the implications.
To summarize, every ratings company fails to obtain a true cross section of the population. However, the leader, Arbitron, with its multistage requirements for respondent cooperation (initial telephone contact, attaining cooperation with mail placement, followed by the need for further cooperation to attain diary return) and with the difficulties posed by its use of multiple diary keepers in the same household is, in my opinion, the least accurate of the lot.
Drawing Constructive Conclusions from Ratings
One ratings book from a single survey period is, to say the least, inconclusive. It’s best to average two or more books’ data together, when there has been no format change, to increase the sample size and your level of confidence. Because the ratings have a definite impact on station revenue and often on the future direction of the station, it’s very desirable to have regular access to more than one ratings service to obtain corroborative and contrasting data. If simultaneous Arbitron and Willhight surveys generally agree with each other, for example, then the reported programming trends and sales data may well be right. If they disagree, you not only have a warning about taking either survey too seriously, but you’ve doubled the chances of getting helpful sales data to maintain station revenue.
Of course, ratings aren’t meaningless. You can learn from them. So now that you understand the limitations and problems of radio surveys, let’s take a look at how you can use them for programming purposes.
The first and most fundamental thing to understand is that the cume figures for the station ought to be much more accurate than the share information. The cume represents an extraction from the whole survey’s data, but the share is simply a mathematical derivation from the turnover ratio between cume and average listening span. (The share calculation is the reverse of the calculation shown earlier. You can use it to restore the data to an approximation of what the ratings company used to create the share data.)
What, then, is your station’s cume in the demographics that you have defined as your target? How does it compare with the cume of competitive stations? Is there any trend visible in the cume over time? If the cume departs sharply from that given in the ratings books that preceded the present report and if there has been no major programming change that might account for it, the change is most likely a fluke.
The cume should be much less volatile than the share. The share is based on a smaller segment of the whole survey sample than the cume, and as an average, it is vulnerable to slight variations in the number of long-listening-span listeners within the sample. So, when there’s no format or competitive change to explain it, a major change in the cume with no change in the share—or with a reverse change in the share—is almost certainly a fluke.
If the station changes its appeal, its listeners will listen for longer or shorter periods of time and will tune in more or less often, thus bringing about a change in the average listening span. However, cume is based on habitual listening. Only if the station completely changes its format will the “tuning-in habit” (cume) change quickly—and sometimes not even then.
Furthermore, if the cume changes sharply in one ratings book and the ratio of share to cume remains the same as before, this is almost certainly a fluke, and the cume should return to its normal level next time. The share drop caused by the cume loss should reverse next time as well.
If there is no station or market change and the cume drops but the share holds steady—or if the cume rises and the share holds steady or drops—chances are you are still looking at a fluke. When there have been no changes to account for it, the cume and average listening spans usually don’t move in opposite directions. I’ve noted many times, though, the tendency of a cume fluke to be balanced by a fluke in the opposite direction in average listening span, so if the share figure shows significantly less change than the cume, the cume change should be an aberration.
All of this is nice to know when the cume (and share) have dropped. However, you must face the same probability of a fluke (and warn your general manager of it) when the cume (and thus share) increases suddenly and for no obvious reason! It’s tempting to believe those “up” wobbles, but they usually lead to devastating drops back to normal levels in the next book—a disappointment that can easily destabilize the station, its staff, and your job, even though it’s just a case of everything returning to normal levels.
Responding to Real Changes in Your Ratings
Up until now, most of my tips about the ratings have been focused on not reacting to a very bad (or very good) book. When can the ratings actually be telling you something real about a change in your audience? When you see a long-term trend in three or more ratings books from the same ratings company.
If the cume is relatively steady and the share figure is up or down significantly in one book, it could be a fluke in the average listening span data, but there is certainly the possibility that something is making the station more or less appealing to its core audience or that a significant competitor is cutting into the listening time of your cume. Whether such a share change really is significant won’t start to become clear until a trend can be identified, and that requires at least one additional ratings report—and preferably at least two.
It’s important to avoid modifying your programming because of a single bad book, even if the ratings drop undermines the sales effort and your general manager calls for immediate changes. If the audience change is simply statistical—a wobble in the ratings data, which the law of probability suggests is likely to occur regularly in radio ratings—any change you institute on the air in reaction to it will needlessly counter established listener expectations, which will tend to reduce the listening spans for real.
As I pointed out earlier in the book, you maximize average listening spans by maximizing frequency of “tune-in” and the length of time spent listening, and that happens when you clearly establish—and then consistently meet—listener expectations. So, even a good change in your programming can have the short-term effect of reducing your share figure until your cume gets used to it! Make programming changes only carefully and thoughtfully.
If a cume trend and/or a share trend upward or downward extends for three books in a row, it almost certainly is genuine.
By the way, if there is a programming change or a special promotion that leads to an upward-trending cume, expect to see your share figure actually drop at first because many of the new listeners will consider your station something other than a first or even second choice. The infrequent listening by your new cume will pull down your average listening span. The increase in cume may not fully offset the decrease in average listening span and thus may depress the share, at least until some of the new listeners change their habits and become regular listeners.
Because of the industry’s tendency to concentrate entirely on share, programmers of mass-appeal stations are tempted to try to decrease music repetition and interruptions, hoping to reduce “irritation factors,” extend their listening spans, and thus increase their share—but there is a danger in this. If the cume is going up or is large, the station is already appealing to a large number of people. That is, it is meeting their expectations, even if some actually prefer other stations; and even those marginal listeners tune it in (when they do listen) based on these expectations.
To be blunt about it, programmers hoping to extend listening spans on mass-appeal stations by installing longer music sweeps, by reducing identifiable station talk and jingle elements, and otherwise streamlining their stations to remove what they perceive as obstructions to longer listening are often removing the very elements that identify the station for its listeners and form the basis for the clearest audience expectations. If they’re lucky, they just hurt their share but retain their cume. If they’re not lucky, the share stays the same or increases, and the cume goes down. With a smaller cume, the share data—even if the numbers are larger—are less stable and more subject to wobbles, and of course, for sales purposes, the station’s “circulation” has declined.
Here’s an example. If a country music listener tunes in a pop station for its news coverage, its personalities, its energy level, or for a change of pace, that country listener is counted in the pop station’s cume, but his or her intermittent listening reduces the pop station’s average listening span and affects its share figure negatively—all resulting from the desirable gain of this country listener. Reducing the number of newscasts, the number of opportunities for the personalities to perform, or the energetic (intrusive) elements of the pop station, will cause this fringe listener to stop listening, instead of extending his or her listening span.
When the share is trending down but the cume is relatively steady, your audience is telling you that their expectations are not being met but they still like the station and hope it may yet live up to their expectations. If you change the overall thrust of the station, you will extinguish their expectations altogether, and the cume will follow the share downward. Instead, do some research, and identify your listeners’ expectations. The changes you make should be toward responding to those established listener expectations; if you make your station what your listeners expect it to be, the cume will remain steady as the share starts to go up again. (The audience will be listening more often and for longer periods of time.)
One situation in which listening span just won’t increase much no matter what you do to extend it is when the station derives a large percentage of its listening from in-car use. When much of your audience listens in the car, you cannot expect that they will drive more often or longer distances just to listen to your station. When they arrive, they turn off the radio and get out. Such stations must concentrate on building and holding their cume because when the listening span is inflexible, the only way to increase the average quarter hour share is to increase the size of the cume on which it’s based.
A station with a large cume and a limited share due to this kind of audience usage should be selling with cume (“circulation”). If the average listening span is shorter than for other stations in the market that have smaller cume, that just means that the advertisers are going to have to buy more ads to reach your station’s great big cume.
In fact, for local sales, if ratings are used at all, the cumes should be the only numbers used because they are the only figures that can be correlated with newspaper circulation figures.
As a former general manager and salesperson, I strongly suggest that ratings not be used for local sales, because what should matter most to a retailer is results, and numerical abstractions from inadequate mathematical studies do not necessarily relate to results. If a salesperson shows “average quarter hour persons” figures to a retailer, the merchant naturally compares them to newspaper “circulation persons,” and the radio buy will look ineffective. However, when radio cume is compared to newspaper circulation, the two media look much more comparable.
To sum up, the best way to study ratings for programming purposes—in which your goal is to learn what is really happening in the market, not just what looks good for sales this time—the key is the cume. Does it wobble? Is it steady? Is it large or small?
In my ratings studies, I rank the stations by the cume in the demographic and 12 + (because, like it or not, the 12 + cumes are based on the entire ratings sample and thus should be the most stable). Then, for each station, I do the calculation I mentioned earlier in this chapter. I divide the “cume persons” into the “share persons” to get a fraction and multiply that by the minutes in the entire daypart being studied, to come up with the average listening span.
This listening-span figure inevitably varies from what it must have been in the original sampled data, because the ratings company has rounded any listening of five minutes or more in each quarter hour up to one full quarter hour of listening. But all stations in the book experience the same rounding, so the ratings data are nonetheless put back into perspective. Do make these calculations; if you only look at the share tables, you really have no idea what is actually happening to your audience—or whether anything is happening to your audience.
When you rank by cume, it becomes obvious which are the niche stations and which are the mass-appeal stations, and with these calculations the variation of listening spans for each type of station and the trending of these spans also become clear and understandable. Only when you understand what the ratings are actually telling you about your station and others can you make informed decisions about making changes in programming that will ultimately improve the station’s position in the community and its billing. Only then can you be sufficiently informed to resist with logic the emotional demands of your superiors to make programming changes that are clearly unwise and counterproductive.