Does Television Advertising Influence Online Search?

>>Professor Ken Wilbur is, well, he’s a professor
right now in Duke School of Business. I think he does a lot of research in interaction between
the advertising technology and media. So I’m hoping this talk would be very relevant to
us since we have a Google TV. We have TV advertising team and also we have, you know, the search
so hopefully, you guys, you know, will be a benefit for this talk.
>>WILBUR: Thanks, Chang. It’s always an honor to be invited anywhere but to be invited back
is a real treat, so, exciting place–you know, everybody is working across disciplinary boundaries.
My background is in econometrics, I work in marketing. I have papers with CS folks and
a few other fields also, so it’s always a treat to come over here and talk to you guys.
Early in my career, I started teaching advertising and I learned about this technology that was
invented in the 1970s, it imperceptively sped up radio advertisements. And so when this
technology came out, there were questions about whether this was good for the effectiveness
of these advertisements and they found that the faster they played the speech, the more
people liked it and the better they retained the information. So my approach has, ever
since I read that, my approach is to talk–it’s always been to speak quickly and I trust that
you will inform me if I flub my words or if you have any questions. So I’m going to spend
about probably about 45 minutes on the main paper here and reserve the last 15 minutes
for something which is a little bit less developed but I think might lead to some very interesting
conversations and maybe give you a different perspective. This one, I think, is going to–I
wouldn’t have written the paper if the answer to that question wasn’t something that looks
like yes. But the second one, I think, is perhaps a little bit more counter intuitive.
So the rate at which people are multitasking has sort of risen toward the heavens in the
last 10 years. So a study done by Nielsen for Yahoo last year said, about 38% of Americans
are multitasking with the TV and the internet every single day. Some academic evidence based
on self-reported media usage from a diary sample, from ’01 to ’04 self-reported multitasking
rose 72% in the span of three years which is kind of jaw-dropping. So this is important
for marketing but there hasn’t been a lot of work in marketing on why that is. You guys
probably have better data than this but this is something that motivated us to study this
topic. We know, in marketing, from–so the question was whether this has continued to
go up a lot. I’m going to take for granted that it has–I don’t have–you probably have
much better data on that than I do. I’m just going to use that fact as motivation for studying
this topic. We know from decades of research and marketing advertising influences consumer
search. So, just cherry picking, a couple of examples from people who worked with me
at Duke, Newman and Staelin found that if I’ve seen an ad then the brand for which–the
brand those advertised, I can recall it much more easily. So if I’m choosing a keyword
to search in a search engine, I might choose the brand rather a category where it occurred.
If I have a little bit of prior information then I tend to search more than if I have
no prior information or a lot of prior information. And so if TV advertising provides some information
that might influence how much I search at a search engine and advertising also influences
how satisfied I am with the results of my search and with the product consumption choices
that I make, and so that–you could make an argument that could map on to the likelihood
I click a result which might be an indicator of how satisfied I am with the search that
I performed. So we’re going to estimate a descriptive econometric model trying to answer
the question of whether television advertising is correlated with the tendency to search,
choice of branded versus generic keywords and probability of click. I’m going to try
to be very careful in my statements and make it clear that this is purely descriptive work
for measuring correlations and associations. I’m going to show you what assumption you
would need to make to interpret these results as causal. I think a reasonable person could
make that interpretation but we’re going to stay away from that ourselves. Why these matters?
For practice, it’s really important in ways that you guys might not like. So if you think
that everything on the search engine comes from the search engine and from consumer actions
taken at the search engine, then you might misattribute some of the effects that are
going on. So if television advertising influences consumer search behavior then if you don’t
take that into account when you plan your TV and search campaigns, as marketers typically
do not, my best guest is in most categories that’s going to lead you to spend too much
on search and too little on television because you will over attribute your sales to search
and under-attribute your sales to television. But then again it’s only a matter of time
before you’re selling out the television ads too. So, it matters for a whole host of other
reasons in practice; what keywords you buy, what copy to put in the ads, when you are
airing your TV ads, how you target to get your audience, what effective frequency curve
you’re using to select programs on TV; just as a whole hosts of reasons this is important.
To the best of my knowledge, it hasn’t been studied in a serious way by academics. There
is some practitioner literature, most of that is either based on self-reports or it’s regressing
quantity of branded searches on timing of TV advertisements, which doesn’t allow you
to disentangle the incidents of search in the product category versus the choice of
the keyword; which is what we’re going to try to tease apart here. For academic research
which is what I’m paid to publish, this is really important because the literature on
search advertising has absolutely exploded in the last five years and as far as I’m aware
with one exception, every single study assumes that the outcomes of the search engine are
determined by elements either that they’re intrinsic to consumer or determined within
the search engine market; and we’re suggesting there’s another way for advertisers to compete
which is through other media. I want to be really clear upfront that the results I’m
going to show you, I do not think they are likely to hold in every product category and
they very well may not hold for different populations of users. This talk is going to
be focused a little more around methodology, so I think that we’re nailing the results
in the setting that we study but I would expect a very different result in other settings.
So we’re going to be looking at the financial services product category. We’re going to
be–I’ll tell you more about the search data in a minute. There’s a couple of reasons why
we’re focusing on financial services. First of all, it advertised on television a lot.
It was the seventh most advertised product category on TV. And we need the TV advertising
because we can observe at the minute and second when those ads appeared and how much we’ll
spent on them and that temporal disaggregation gives us a lot of power to identify the effects.
Secondly, the brands in this category have relatively unique names. So Charles Schwab,
E-trade, Ameritrade even words like fidelity are not often used another context. If I had
studied something like computers then it would be tough to separate the searches for Apple
from the searches for the fruit, right? So this brand name overlap with common words
and names problem is surprisingly frequent when you look across product categories. Third,
this category needs to be something that people actually search for. So if I was doing this
on cola, chances are people don’t search that much for cola because it’s really cheap to
buy a cola, try it out and get your information about the products that way. So this is a
high-involvement category where people don’t purchase very frequently suggesting that they
probably spend a lot of cognitive resources to search for the product that they choose.
Fourth, simultaneity is going to be really important in this study because, at one point,
I thought about doing movies. So movies fit these other criteria pretty well. The problems
that TV advertising for movies peaks around the date of their release as does search activity
for movies and so is the TV advertising causing the search or is it the search causing it?
It would be really hard to tease those apart. In financial services, it’s not quite so suspect
in terms of, is the TV advertising being caused by the search activity. I’ll go into that
in a little more depth in a minute. Finally, we’ve got some time-varying data which will
allow us to isolate the effects of TV advertising from other factors that might be causing search
and we’re going to use the stock market index which is why they reported in the media as
an instrument to tease out these effects. So the online search data; we’re not advertising
the source of these data but you can probably figure it out, so what we’ve got is about
20 million queries for randomly selected sub-sample of several hundred thousand users of a top
ten general interest search engine. It’s not Google, Yahoo or Bing. It covers three months
in 2006, for each query, we see an anonymous ID. The term that was searched–they didn’t
time the search–if a result was clicked then we know the position of the result and we
know the URL that was clicked. I should mention we’ve been very careful to preserve the anonymity
of the user data by analyzing this data in the aggregate in order to comply with our
institutional research policies. So what we wanted to do was identify a bunch of branded
keywords and a bunch of product category-related generic keywords and we didn’t think that
we had a complete list of those when we started the project. So we developed an algorithm
to do that. You guys probably have a better one. I don’t think it’s been published anywhere
but if that’s incorrect, I’d love to know about it. So just to sum it up and the details
were in the paper, what we do is we look at keywords that lead primarily to click someone
brands website and we’re going to call those keywords that are related to that brand. Keywords
that lead to clicks on any brands website we’re going to call generic and keywords that
often lead to clicks outside the category we’re going to call unrelated. So a couple
of examples, things we didn’t expect to find but we did find and have a high face validity,
some branded keywords includes some very common misspellings of the brand names Charles Schwab
and fidelity. I no longer know how to spell Charles Schwab because I’ve looked at the
misspellings so many times. There are some brand keywords we didn’t expect to see. Generic
keywords, all–you know, look pretty reasonable. A couple there, I didn’t expect UGMA turns
out to be a particular type of intergenerational wealth transfer. It’s a trust fund. It’s a
tax shelter, basically. So things that I didn’t expect to see but after they came out made
a lot of sense. And then words that were often appearing in queries with these other words
but we’re not particularly related to financial services that we eliminated, like calculator
and government and education and so forth. If there’s a particular word that led to–that
led to clicks on in a high percentage of queries where clicks occurred, I’m going
to assign that word to fidelity. If that word led to eight different brand websites on a
high percentage of queries where any result was clicked, I’m going to call it generic.
And if that word led to financial services brand websites at a relatively low rate on
all the result clicks for the queries that included that word, I’m going to call it unrelated.
So it worked a lot better than we expected it to. We think that this could be useful
in a lot of different settings like if you’re trying to identify particular keywords within,
say, a Twitter database or–my favorite application if I was in a law firm doing discovery on
a giant lawsuit or I’ve been delivered 20 gigabytes of emails and had to find which
of those ten emails were the smoking gun–you could do something similar to this. So the
TV advertising data is fantastic. If you’ve never heard with Kantar, they’re great. For
every brand and a product category we’re going to see the minute and second at which their
advertising appeared and estimated cost of the advertisement and then we did a lot of
analysis of the advertising content. And the two variables we settled on using or calling
web emphasis and phone emphasis. What that means is that they either said the name of
the website during the commercial or they said the phone number during the commercial.
They both correlate pretty highly with other measures we thought of to measure the same
thing and they were very easy to objectively quantify. We’re going to–because our purpose
is here descriptive, we’re going to define four user segments and all of our parameters
are going to vary within each of those segments. Because we don’t have demographics, we can
only segment based on user graphics. So if you searched more non-financial services queries
than the median consumer–I’m going to call you a frequent searcher and if you searched
anything within the past hour, I’m going to call you a recent searcher. Of course, we
only have data for the one search engine so there is some risk of misclassification of
users here, but overall, I think it works pretty well. So to explain each of these behaviors,
we’re going to have brand fixed-effects. We’re going to control for all of the temporal variation
that we can in terms of the week of the sample, hour within the week day, and hour within
the weekend. Age of the brand; which correlates highly with total assets under management
as well as recent asset growth under management. And, we‘ll use the changes–the daily changes
in the stock index and hourly changes within business hours since those are widely reported
in the media. This is mid 2006, 7% of advertising expenditures met are definition of web emphasis
which means they said the name of the website in the ad. And 16% of these expenditures said
the telephone number within the ad. So the brands were emphasizing telephone contact
much more than they were emphasizing web contact and the same pattern shows up in a variety
of alternate definitions of web emphasis and phone emphasis. We’re seeing people search
more on weekdays–which, in this category, makes sense–most of these people are probably
professionals. They probably do more searching in the office than they do at home but we’re
seeing much more advertising on weekends. I’m going to come back to these in just a
So if I graph the advertising expenditures and the total category searches by date within
the three-month sample. I’ve got the law–I’m sorry, the advertising expenditures deflated
by a factor of a thousand in the solid line, total category searches and the dashed line,
it’s hard to see any correlation there. In fact, it’s a slightly negative correlation
because the ad expenditures spike on the weekends and the category searches drop on the weekend
days. Yes. So we experiment a lot of different definitions of click-through rate and the
one we settled on was whether you clicked. So if you searched the brand name at least
once within an hour, then we’re going to say that you searched the brand. And if you clicked
at least one brand-related website, then we’re going to call that a click-through rate. So
the search engine that provided the data did not display many paid links. When they did,
they were lightly-shaded and otherwise, similar in appearance to the organic links. They typically
displayed either zero, one or two paid links. On the branded searches–branded keyword searches
which is what we’re focusing on–those paid links were almost always the same as the top
organic links. And so we don’t have information on clicks on ads but in this context, we’re
not missing a lot. We also tried to get website traffic data. Alexa will not sell that to
you if the brand has paid them to withhold it; which was the case. We also tried–yeah,
we tried to supplement the data in few other ways to the best we could. So if we look at
the same pattern across hours within weekdays and hours within weekends, we again see close
to a lack of correlation in the two variables and so this data suggests to me that the brands
were not planning their TV spending based on the expected search habits of consumers.
So if you make that assumption which is only an assumption then you can interpret the rest
of the results as causal. I don’t have hard evidence of it either way, so I’ll try to
stay away from any causality claims. Although I did present this at work and there was a
guy who worked at Fidelity at the time this data were produced and I asked him, “Did you
plan your TV expenditures based on user search behavior?” And he said, “Absolutely, not.”
But again that’s only one data point. So, okay. So we’ve got three models, we’ve got
four segments. We’re going to relate all of these behaviors to each other using correlations
in the error structure.
And I’m going to go kind of quick, just tell me if you have any questions about anything.
For advertising, we’re using exponential smoothing assumption where basically the–the second
equation there, the advertising stock at any–in any hour of any day and the sample is going
to be just as smooth function of recent advertising expenditures. So this is for searching the
category. What we’re controlling for here is gamma one is a segment-specific tendency
to search any word in this category. We’re going to let beta one which is the coefficient
on advertising, we’re going to let that vary according to week day, hour and week of the
sample. We’ve got the stock index in there. The XT terms allow the search incidents to
vary with time effects and we’ve got an error shock at the end which I usually called Zi
but some people called Cazi and some people called Casi. And I’m going to say Zi; which
I’ll show you how that place in later. So on the keyword choice, conditional on having
searched in the category, what type of word do you choose, we’re going to lump all of
the branded keywords within our brand index by K into one option, because what we saw
in the data is that if you searched a keyword related to a particular brand, you almost
never search a keyword related to any other brand within the same hour. And if you searched
a brand-related keyword, you almost never went back to a generic keyword within the
same hour. So we’ve got a similar modeling structure predicting the tendency to search
the keyword. Again, the baselines are brand-specific in this model and then the advertising responsiveness
parameter also depends on the characteristics of the ad such as weather emphasized web contact,
phone contact and the age of the brand. Finally, conditional and having searched the any brand
related keyword; if you click at least one result then we’re going to model that with
the binary legit with similar predictors. In terms of the air correlations, those Cazis–what
did I say? Zis. The Zis, we’re going to allow them to be correlator across models, across
segments, across time and we are going to choose parameters to maximize the stimulated
log likelihood of all of the data at once. And, I’ll be happy to go into details later
if you want me to. They’re kind of boring, they’re pretty straightforward. The only thing
that we really learned during, from a strictly modeling point of view that we didn’t know
going in was–we didn’t appreciate the importance of waiting for–how do I say it? So in–say
at 4 o’clock, a thousand category searches occurred and at 5 o’clock, 800 category searches
occurred; then we needed to weight the amount of information in the keyword choice model
within each time period according to the number of queries that occurred at each time period.
So, in retrospect it’s pretty obvious but going in we haven’t expected that. It’s–so
this is the keyword choice. So basically what we’ve got is a multinomial distribution were
C in the bottom equation, C is the number of category searches that occurred. C1 is
the number of category searches that occurred and C2 is the number of searches for brand
K. So, I wouldn’t, I–it’s closer to important sampling than to normalization. Okay. So,
how does the model perform? Fairly well, R square around .7, we do a holdout sample and
it’s–even a nonrandom split, so we estimate the model based on March and April, we predict
for May and we see how well we do. The errors in May are less than 5% or typically about
5 to 6% higher than the errors in March and April. So that’s a pretty good performance.
What do we find search behavior? We don’t find any significant correlation between TV
advertising and the tendency in the search in the category. This is why I threw these
caveats in at the beginning of the talk. If this was a new product category, if there
are a bunch of new brands in it or if it was still evolving, I would expect to find a different
pattern of results with regard to search behavior. In this category, I think it makes sense.
Basically, what this says is that, if I see a TV advertisement, I am not likely to–I
am not more likely to search any financial services brand that search for financial services
are–is driven more by, maybe lifestyle or demographic or employment or financial factors;
not by TV advertising. You might imagine that something different in another setting.
>>Do you find the volatility of the market actually has the most impact on search behavior
items?>>WILBUR: Yes. We have a very similar result–well,
we have some similar results that people are searching based on how the market is doing
recently. Yeah. We are finding that TV advertising correlates with choice of a branded keyword.
So, if you were to take the causal interpretation then–if I see a TV ad, I’m less likely to
search a generic keyword I’m more likely search a branded keyword that has tremendous implications
for the marketer regarding the breath of competitive information available to the consumer and
also regarding the likely costs of paid clicks on the search results that the marketer has
to pay. Those effects are biggest for consumers who search often and consumers who have not
searched recently. So if you take those together, it’s just that the dominant mechanism might
be, that if I am a frequent searcher and I see a TV ad, I haven’t searched recently but
I will start a search rather than, you know, I’m already online, I hear a TV ad in the
background and that influences my search behavior. So it suggests TV advertising more likely
to start me multitasking rather than effect me while I am multitasking. The effects are
biggest during business hours and for younger brands. We measure the effects in terms of
elasticities and we’re finding that they’re pretty comparable to other studies; measurements
of shorter on elasticities of TV advertising on sales. So, that’s for younger brands. So,
somewhere around .05, .06, .07. Not big but not smaller than other measurements of advertising
effectiveness. In terms of clicking, we’re not finding a lot of action. Some consumers
are more likely they click when older brands advertise on TV and people who search a lot
clicks slightly less when brands advertise on TV. That might suggest that some of the–some
of the brand queries that do the advertising is associated with–are less likely to convert
than in the absence of TV advertising. So, in terms of the duration of the effect is
basically in hours, not days and it last longer for people who search frequently. I want to
be really clear, the boundary around the study, there hasn’t–in the academic world as far
as I’m aware, there’s been no work on this topic although I’m sure there’s been work
here internally and I’ve already heard a couple of whispers about that this morning. We’re
only looking at a particular product category that behavior in 2006 is probably very different
from the behavior today. It’s a mature category; I would expect different patterns in a new
or evolving category. We’re relying on time-series identification and really we would prefer
single source panel data. But, hopefully, we’ll spur some additional research on this
topic and hopefully we’ll encourage marketers to start planning these two ends of the purchase
funnel holistically, maybe we use the same agencies to, you know, work on both sides,
maybe adjust ROI metrics to think about how expenditures in one medium impact outcomes
in another. And also for academics we–I would argue, it’s very important to consider the
possibility that, you know, my brand A’s click through may be inherently much lower
than brand B’s but that doesn’t mean brand A is defenseless, they can take action–they
can take a variety of other actions. So, I mentioned that I wanted to save about 15 minutes
for the more formative study, before I transition over to that, are there any comments or feedback
on this one?>>No. I was just going to say that your intuition
a few minutes ago is true that we’ve done working this area and do–and do see a new
searches from TV advertising. I think for some of the reasons that you’ve mentioned,
financial service is not a great example, although we do–even in financial services,
we do see–we do see an impact. That’s all I want to say.
>>WILBUR: So let me follow that up with a question to you. So, are you guys–are you
thinking about separating the category search incidence from the keyword choice or the choice
of a branded keyword?>>So I don’t–so I don’t know that we’ve
done that specifically, I have to think about that. But it’s a good point, yes. And I think–I
mean, so you said very early on, is definitely saying that we found to which is, you know,
different categories have, you know, a different kind of draws for people to go online. I mean
we find financial services to be one of the better ones in part because there’s lot of
information online for that. And as you said things like consumer products are relatively
poor because it’s just–the likely to find so interesting online is low and the ease
of getting information by just going and trying it out is high, so.
>>WILBUR: Yeah. And a lot of these things are actually delivered online, and I would
expect different behavior. If I–I might even expect different behavior if I looked at a
different search engine’s database…>>So why did you use 2006 data? Is that just
the only data you could get?>>WILBUR: I’m a poor academic–yes, that’s
the only data I could get.>>But because the financial market [INDISTINCT]
me into like, in 2008 and 9–if you redo the study, even the data, you think it’s probably
a lot noisier or…?>>WILBUR: It’s a good question. I don’t have
any feel for that. Yes?>>Quick question. You mentioned earlier in
your presentation that you looked at advertisement that call fraction on the websites or on the
phone number.>>WILBUR: Okay.
>>Did you see different–different user behavior in one case or another?
>>WILBUR: So one of the–more surprising results to me from the study is that we didn’t
find any significant correlations with web emphasis within the ad itself. I suspect the
reason for that is that we did not have website traffic data and so it–maybe that people
just want directly to the website when the website was said during the ad. And I–I get
in trouble when I talk off the top of my head but I think we were finding that advertising
for older brands actually lowers click through rates for some consumers and we interpret
that as indicating people maybe more likely to call on the phone rather than search online
or click through on a search. So with that, I’m going to–I’m going to keep it brief but
I’ve got–we’ve got what we think is a bit of a provocative result and I just kind of
want to throw it out there because I think it might generate some interesting insights
in it. I think it’s relevant to some of your core business. So, the lead author on this
paper is Linli Xu, she’s a very talented doctor student at University of Southern California.
And, what we’re looking at is price advertising in particular. So I’m going to start with
three facts that I think nobody can dispute. The first is that, price is often the signal
of quality. So, if–this was shown in market equilibrium by a couple of very famous economists,
Milgrom and Roberts, and this has been tested in the lab dozens of times by dozens of studies;
people infer product quality from the price that they are quoted. So, I’m going to take
that as fact. Second fact is that, marketers are often using advertising to communicate
price saying this might be cheap or this might be luxury or this might be high-end and I
don’t have a citation other than three very famous co-authors DUH. And finally, the proceeds
source of an advertisement is important. So, I used to teach advertising, I use the leading
advertising textbook; this textbook identified three factors of an ad as being critical as
to determining its effect, that is, the perceived source along with the message and the medium.
So, when I say source, it’s whenever you receive a communication, who you perceive that communication
to be from; it could be a celebrity spokesman within the ad but just as often it’s from
the organization that paid for the ad. So, if we start from those three facts, then we’re
going to come to a prediction which is going to be the central theme of this paper. That
price advertising by the manufacture should be less effective than price advertising by
dealers or retailers. The reason for that is the manufacturers are responsible for the
product quality. And so if I get a price ad from manufacturer, there’s a credibility problem.
If your product is so good, why is it so cheap? If I get the same price ad from a retailer
or a dealer, there is no such credibility problem. I can attribute the low price to
inter-retailer competition or inventory concerns but I’m not likely to make the same quality
attribution to the retailer or the dealer because they didn’t make the product themselves.
Okay? So that’s going to be out prediction. We modified our price advertisement to test
this prediction experimentally. And we’re looking at the automobile case because in
that industry we see a great deal of price advertising by both retailers and manufacturers.
So, I’m not going to be able to show you the actual stimulus but I can describe it for
you pretty quickly. So we took an ad from Ford, the script of the ad was, “Since we
introduced the Ford family plan, hundreds of thousands of Americans have joined in on
the savings. Now, we’re extending the plan until next September 5th. Get employee pricing
on that series. America’s best selling truck, blah, blah.” We took that ad, we stripped
out the audio, we played a Toby Keith song in the background and I read the same script.
In that way, we can have the same voice and music in the two ads. In the other condition,
we did–we made three changes. All of which are consistent with industry practice. First
of all we changed the subjects and the pronouns in the script, so rather than, “Since we introduced
the Ford Family Plan,” “Since Ford introduced the Ford Family Plan.” So this time, we’re
giving the impression that the source is the Texas Ford Dealers Association. “Hundreds
of thousands of Texans in have joined in on the savings.” That’s the other change we made.
We changed the geographic frame of reference from the national market to the Texas market
because that’s what local dealers’ associations typically advertise. And third, we changed
the logo in the final three seconds of the ad from Ford to the Texas Dealers Association.
So, in some there was a pretty subtle manipulation we have over 90% of the same words in the
two scripts and over 90% of the same video frames in the two ads. We first did a pretest
with a 150 people; we showed them one of the two ads at random. We asked who paid for it,
was it Ford or was it the Texas Dealers Association or do you not know or someone else? And we
found that people got it right on average. Almost 80% of people got it right in each
of the two ads. Next, what we did was created a survey, so each respondent came to the website.
They saw a Ford brand ad showing how tough the truck is and then they got one of this
two price ads at random. Then we ask them a series of questions, we said, indicate your
agreement on a 1-7 scale with each of the following questions: Ford trucks are a good
value; Ford trucks are high quality; Ford trucks are tough–that’s the branding claim–I
would be likely to test drive a Ford truck; I would be likely to purchase a Ford truck.
And then we asked what types of truck they had owned in the past. And we advertised the
survey on Facebook to people who had indicated in their profiles that they like pickup trucks
or that they like particular models of pickup trucks and a couple of bulletin boards. So
these are the results. This is a sample of about 350 pick-up truck enthusiasts or consumers.
The dark bars are those who were exposed to the manufacturer price ad, the light bars
are those who saw the dealer’s price ad. And all of the differences are significant with
a 95% confidence level. And it’s a 1 to 10 scale. So I’m asking you your agreement on
a 1 to 10 scale with each of those statements: Ford trucks for high quality, good value;
Ford trucks are tough, et cetera. So all we did was change the perceived source of the
price ad from Ford to the Texas Ford Dealers Association, and on a 1-10 scale we see a
bump of about three quarters of a point in the perceived quality of the truck. It’s a
really big effect, surprisingly big. And we took the ad that existed in the market place
and it was in our altered condition, the dealer’s price source that we’re doing much better.
It’s not like we changed it to make it worse, we changed it to make it much better. We also
asked people who they thought paid for the ad, whether it was Ford or the Texas Ford
Dealers Association or someone else. So, if this affect is occurring and people are making
this inference, then it should be the people who thought Ford paid for it have the lower
quality perceptions. And that’s not random assignment, it’s–we’re based on–it’s based
on reported attribution but we’re seeing almost exactly the same pattern of effects. So the
dark bars are people who thought Ford paid for the price ad, the light bars are people
who thought Ford–the dealers association paid for the price ad. Almost the same size
of the effects, first four significant of the 95% level and the last one significant
of the 90% level. That might be my last slide, it is. So, this is–we do more in the paper.
We uncovered a similar pattern of effects using market data for pick-up trucks from
five years of advertising and sales. But this is something that you have going on in your
paid search ads all the time. I see a lot of search ads that say cheap, that say exclusive,
that–I see a lot of retailers advertising on manufacturer brands. And I thought I would
throw this up and–because the first one I knew what was going to come back toward me,
but this one I didn’t know he’s going to come back. So, I wonder if this is something that
you guys have thought about or if you think it might be interesting for your business.
>>I’m just wondering whether you’d controlled for the second variable that you changed in
the [INDISTINCT], where you changed American to Texan.
>>WILBUR: I went through it really quickly, so let me try to clarify. We randomly assign
them to one of the two price ads. After they answered the five questions about quality,
we asked them “Who do you think paid for the price ad?” And so, this is based on
stated attribution, so if they thought that Ford paid for it, they’re going to go into
the dark bar, otherwise, if they thought the dealer’s association paid, they’re going
in to the light bar. So, not everyone got it right, but what we’re showing is that independent
of the random assignment of which ad they got, if they thought Ford paid for it, we’re
seeing the same effect. Now, we think one of the implications of these result is that
when the manufacturer does do price advertising, it probably should not play up the fact that
it’s from the manufacturer. So you don’t want to highlight the fact that the manufacture
is paying for it. You may want to–in the TV context, you might want to use the different
announcer, maybe different agency, maybe a different campaign theme. In the search ad
context, maybe you don’t want to direct consumers to the URL of the manufacturer. Maybe you
set up a retailer under a separate brand that’s owned by the manufacturer and direct consumers
there if the manufacturer’s running price ads. Okay. Not quite as provocative as I thought.
>>Last question here.>>WILBUR: Okay.
>>Do you think it’s really difference between manufacturer and dealers or it’s just because
the consumers perceive a local endorsement of the dealer association? I mean, it’s not–it’s
something to do a low cap–local community or whatever that perception is something you
can tease out, yeah.>>WILBUR: That’s certainly something that
we can tease out and we haven’t focused on doing that so far.
>>Yeah, because it’s kind of relevant to the paid advertising because a lot of times
our ads, you know, people in the creative, they want to add a local location and, you
know, it’s tied to your city, to your town. This seems maybe, you know, will improve the
click-to rates. It’s just one of the tactics some people might use, yeah.
>>WILBUR: I completely agree.>>Yeah.
>>WILBUR: Thanks.

Related Posts

How to Find Backlinks from Competitors site । SEO Backlinks checker tool । How to check Backlinks

How to Find Backlinks from Competitors site - SEO Backlinks checker tool - How to check Backlinks please subscribe my
Link Building Strategies on Steroids: How to Get Backlinks FAST!

Link Building Strategies on Steroids: How to Get Backlinks FAST!

In this video, I’m going to show you how to turn your backlink analysis into actionable link building strategies… fast.

Leave a Reply

Your email address will not be published. Required fields are marked *