U.S. advertising expenditure, for various media and type categories across the years 1919 to 2007, is now available in a dataset convenient for extensive analysis. These data quantify the rise of advertising on radio, on television, in telephone directories (yellowbook), and on the Internet. They also quantify less widely discussed media for advertising, such as direct mail, billboards and outdoor advertising, and advertising in business papers (trade press). You can, of course, also create your own categories, such as ad expenditure totals relative to GDP. Advertising is widely regarded as an important aspect of new media businesses. This dataset is meant to inform thinking in this time of dramatic changes in the communications industry.
The advertising expenditure data mainly come from Robert J. Coen, now at Magna in the McCann Erickson advertising agency. Coen has worked for decades putting together advertising expenditure figures. His advertising data were published in the U.S. Census Bureau’s Historical Statistics of the United States, Colonial Times to 1970 (see pages 855-7). The Television Advertising Bureau has made available online a recent version of Coen’s data covering the years 1948 to 2007. I’ve augmented that data with Coen data for 1919 to 1947, previously made publicly available on the Internet. In addition, I’ve added some categories of advertising expenditure for 1919 to 1934. These additional figures are based on some Coen data, some data from other sources, and my own estimates. In recognition of Coen’s work, I’ll call the dataset the “Coen Structured Advertising Expenditure Dataset,” or the “CS Ad Dataset” for short.
The CS Ad Dataset includes various advertising expenditure categories and aggregation schemes. These are described in the CS Ad Dataset “categories” and “schemes” sheets. Separating advertising into national and local advertising is possible from 1935. The category “Television” was divided into “Broadcast TV” and “Cable” in 1990. The category “Out of Home” replaced the category “Billboards” in 2000, with the former encompassing about three times as much expenditure as the later. Categorization has become more specific over time. In 1935, about 20% of advertising expenditure fell within a “Miscellaneous” category. By 2007, the share of “Miscellaneous” was down to 13%. To deal with issues of categorization and aggregation, the CS Ad Datset includes records containing the year, two levels of categorization, an aggregation scheme code, and the associated advertising expenditure for all years and categories from 1919 to 2007. This data structure facilitates analysis using a relational database such as Microsoft Access. If you don’t have such software, you can extract relevant data by re-sorting, cutting, and pasting as best suits your needs.
The quality of the advertising expenditure data merits further analysis. Since Coen has long worked for an advertising agency, Coen’s employer has particular interests that might bias the data. However, Coen has made the advertising expenditure data widely available for a long time. Public exposure greatly increases the credibility of the data.
Some advertising expenditure figures have been revised over time. Between 1975 and 2000, the total advertising expenditure figure for every year prior to 1940 were revised by more than 1%, with the figures for the 1920s revised downward about 15%. Detailed categories changed more that the totals. For example, totals between 1935 and 1945 were revised by less than 1.8%, while there was a -6.3% revision for magazine advertising in 1938, and -5.5% to +8.2% in the miscellaneous advertising figures. Putting aside the most recent year of data, the largest revision to total advertising expenditure for post-WWII years was a 2.5% increase for 1998. This revision occurred between 2003 and 2008. The “revisions” sheet in the CS Dataset shows some revision history for the total figures. Overall, data other than the most recent year appear to be reasonably stable. Nonetheless, since revisions are ongoing, dating the Coen data used is important for consistently defining figures.
The Newspaper Association of America (NAA) lists mainly Coen data on its website under a page-tab “annual ad volume” (all-media table). The NAA, however, inserted its own estimates of advertising expenditure in newspapers, an area of advertising for which the NAA has a particular interest. The NAA newspaper advertising expenditure in its all-media table doesn’t respect the over-all aggregation scheme of the data. Specifically, the NAA newspaper advertising figures include online newspaper advertising, while Coen has a separate category for Internet advertising. For details, see the “NAA Comparison” sheet in the CS Ad Dataset. In addition, while the NAA inserted its own newspaper advertising expenditure figures, it did not change other Coen figures to account for this change. Hence the total figures for the NAA all-media table don’t add up. The NAA table also seems to include different revisions of Coen’s figures. Hence the NAA all-media table is’t a good quality source for consistent, historical advertising expenditure data.
Note: The CS Ad Dataset is available online and as an Excel workbook. See also additional discussion of ad expenditure.
Update: For some additional comparative advertising expenditure data from the U.S. Census Bureau, see U.S. advertising expenditure, 1998-2007. Here’s a review of alternate advertising expenditure data, including IRS data on advertising.
37 thoughts on “U.S. advertising expenditure data”
Holy cow. That’s really interesting and useful information, well presented.
This is an incredibly helpful and valuable project to make available to all of us who work, study, and live in the media and advertising business. Thank you for your time and dedication put into compiling this data. Special thanks as well to extraordinary work of Robert Coen.
I have been searching around for a while trying to find comparison data across channel. This is exactly what I have been looking for — and more. Thanks very much to you and Robert Coen for providing.
Thank you SO much for making this available. I have been scouring the internet for this. You broke everything down into sections, which serves my purposes perfectly.
Thanks so much for doing this. Saves a ton of work.
One nit: the 1911 number on the summary sheet seems to be missing a zero.
On the internet numbers: Coen seems close to the TNS numbers. I’ve been told these are display ad only numbers. The IAB is IAB number is display, search, email, lead generation and classifieds. I think you’re right, though, that changing these would mean decreasing something else. (The IAB, like the NAA, is also not necessarily the most objective source.)
Thanks for the comment. I’ve corrected the 1911 figure: it is 1,010.
Also, I added some online advertising figures from the Service Annual Survey to the internet sheet.
Thank you for these helpful data. However, it does not say anything about Coen’s methodology. Is he doing surveys? Do you know how he actually generates his data?
On Coen’s sources, see
For comparison of Coen’s figures with some government-generated data, see
Thanks for this helpful data! One question: are the categories mutually exclusive? I.e. does the “newspaper” category include internet-based newspapers? Or are those captured entirely by the “Internet” category?
The Coen newspaper category appears to include only print newspapers’ print advertising. The categories in the advertising summary table are mutually exclusive. The full Coen structured dataset includes a complicated set of categories and aggregation schemes.
Here’s some more comparative analysis of newspaper advertising statistics.
Why does this data end in 2007? Looking at old Ad Age magazines, it seems that Coen used to update this frequently. Is there any reason that he stopped?
Coen retired by early 2009. That hardly gave him time to produce data for calendar year 2008. But if you find 2008 data published in Ad Age, please forward it to me and I’ll add it to the dataset.
The best I have found is the graph below which contains some 2009 and 2010 estimates (noted under the graph). I have been looking for data from recent years to analyze the effects of the internet. If I find anything else I will definitely send it along.
By the way, this dataset is a great resource for my research and is very interesting. Thank you for making it available in such a useful format.
This dataset is a great resource! Thank you for putting it together!
A question: the spreadsheet indicates that “grand total” US advertising data is in current dollars. To which year does “current” refer? And how were they indexed (e.g., CPI, ad-pricing, etc.)?
All the figures are in U.S. dollars, not adjusted for inflation. That’s what I meant by current. Thanks for identifying the confusing terminology, which I’ve corrected.
Hi! Thank you for your great work, it’ll help me a lot for my master thesis!
Also, would you happen to know where i can find the same data for the Eurozone?
Sorry, I don’t know where to find Eurozone data. You can find some historical advertising data for the UK in Table 6 of my paper on convergence.
Firstly, great article and thank you for including links to the Coen’s data sets as well as an overview of his methodologies. That info has been a great help! I’m currently researching the variances between Coen’s figures and the figures reported annually by Kantar Media and had a question for you.
In your research of Coen’s data, have you seen notes or comments describing how he defined his subcategories? For example, “Local Newspapers data is based on X# Local Publications” or “Internet is based on X sources”. In your article, it sounds like Coen also added Online Newspaper/Magazine data into his Internet totals, but I can’t seem to find as references for the definitions he used for his categories.
I haven’t seen any definitions of Coen’s categories or sources. Coen worked for ad agencies. His sources seem to have been a variety of industry advertising sources. Looking at the ad industry trade press (e.g. Advertising Age) over time might give you additional insight.
Thanks Douglas, I’ll start there.
THANK YOU SO MUCH FOR THIS DATA! I did all caps on purpose.
Has anyone updated this to the current year?
As far as I know, no one had updated the data. I doubt that it would be feasibly to update it, because many of the media categories and advertising data sources have changed.
It seems the Google Spreadsheet link is no longer working! Could you please repost it? It would be so helpful!
All the links should be working now. If not, please describe more specifically which link isn’t working. The data are also available the data as an Excel workbook. You can open that in Libre Office if you don’t have Excel.
Please let me know the data of Advertisement & Branding Expenditures of School & Colleges in USA.
I don’t know of any advertising data in that narrow category.
Thank you for your excellent work and Coen’s continuous dedication!
I read your series of articles on this topic, and I want to discuss some questions with you:
1) Early data: 1867~1935
Coen has revised the data of earlier years and saw that your subsequent data refer to the latest version of the data. The absolute difference before and after the revision of the number before 1930 was 300 to 500 million, so why did this change happen?
Why did you choose the revised data? Generally speaking, people will choose to use the current data recorded in the past era.
2) Newspaper: There is a big difference in the amount of data from a data source, but overall the difference between NAA and Coen is within 1%, IAB is about 2%, and SAS is very different.
In your article, NAA vs Census, you are more inclined to believe in census, I think the mainstream amount is closer to NAA?
Secondly, the data of IAB in 2006 and 2007 did not decrease but increased. The trend is different from the others, which is very strange.
We don’t know what the IAB’s data sources and statistical methods are, and there is no way to verify it.
3) Internet advertising:
I have a guess. Regarding the iteration of new technologies, the popularity of PCs has impacted traditional print media such as newspapers and magazines, while mobile+3/4/5G has impacted video media (TV).
What do you think about that?
Good that you’re studying these statistics carefully. On Coen’s revision, given his detailed knowledge of the statistics, I think his revision is credible. I think Census Bureau is more credible than NAA because Census isn’t a directly interested party and have more publicly accountability. I think your guess on Internet advertising is smart and insightful.
If you do work on these data, I urge you to share your work publicly. Feel free to include a link to your work here.
Above is my preliminary result.
sheet2: IAB data (1999~2019), including data such as the structure of advertising media, the form of Internet advertising, and advertisers.
Sheet1: Long-term advertising data in the United States (Thank you again for your work), collaged with data from Coen (1935~2002, 2007) and IAB (~2019) to obtain the complete structure from 1935 to 2019.
Through drawing the structure (at the bottom of the sheet), there are some findings:
1) In addition to miscellaneous items, newspapers and television, as the two pillars of the early years, occupied the main share. Television slowly ate the newspaper market.
2) Internet advertising has grown rapidly, so newspapers have accelerated their decline, and TV share has declined (the absolute amount is still stable).
In addition, I am collating advertising industry data in Japan, China, and Germany. If necessary, I can share and discuss them.
Great work! Thanks for your generosity in sharing your work with everyone.
As you noted, the US TV advertising share has declined while its dollar amount has remained stable. From 2000 to 2019, the US advertising share in GDP is U-shape between 1% and 2%. Despite huge changes in advertising technology, there’s no overall trend increase in ad share in GDP. That’s what I found in the US from 1919 to 2007:
Similarly for the UK:
A new type of advertising, not captured in market-based advertising spending, is relationship advertising. Amazon, with its massive data trove on consumer purchasing behavior and powerful machine-learning tech, does relationship advertising through its on-site recommendations and email marketing. Growth in advertising relative to GDP seem to me likely to occur implicitly through relationship advertising/marketing driven by big data and machine learning. That may unfortunately favor more massive business corporations.
Collating advertising industry data for Japan, China, and Germany would be useful and interesting. I encourage you, in order to get more credit for your work, to start your own website and share your important work through your own website. Of course, you’re always welcome to provide useful comments, analysis, data, or links here!
Thanks. Have you considered adding to your data that from 1865 to 1937 published in Neil H. Borden (1942) The economic effects of advertising (Richard Irwin, Inc. p. 48, Table 1)?
I ask, because I’d like to merge your data with Borden’s in an update to the plot of your data that I posted to Wikimedia Commons on 2018-08-31 (https://commons.wikimedia.org/wiki/File:Advertising_as_a_percent_of_Gross_Domestice_Product_in_the_United_States.svg). McChesney and Nichols (2010) The Death and Life of American Journalism (Nation Books, pp. 133-134, 270) plotted Borden’s data, arguing that during that period, the press became “independent of government encumbrances but increasingly … under growing commercial pressures to downplay partisanship”. “Confirmation bias and conflict” on Wikiversity claims that more recently, commercial pressures have increased partisanship (https://en.wikiversity.org/wiki/Confirmation_bias_and_conflict).
My advertising data are annual from 1919; Borden’s data prior to 1919 is newspaper and periods advertising revenue estimates from the Census of Manufactures for the years 1865, 1866, 1867, 1880, 1890, 1900, 1904, 1909, 1914. For 1880 on, you can find much more detailed newspaper and periodicals advertising data and analysis at my post on U.S. newspapers’ revenue structure, 1880-2007. As for the data for 1865, 1866, and 1867, $7,584,340, $9,686,844, and $9,600,327, respectively, I’m skeptical of the value of those aggregate figures.
Thanks for the reply.
What would be your reaction to translating Borden’s numbers into percent of GDP and plotting them with your 1900-1939 and 1919-2007 numbers as three separate lines, with Borden’s being dotted gray, your 1900-1939 being dashed black, and your 1919-2007 being solid blue — or something like that — with a caption explaining the source and indicating that the older numbers are less reliable? Even if the numbers aren’t great, they still to me seem useful for indicating the general trend as advertising became the dominant source of funding for the news media, thereby giving the media a conflict of interest in reporting anything that might offend a major advertiser, as discussed in Wikiversity, “Confirmation bias and conflict”. Or maybe show four lines, not three, being (1) Borden’s, (2) your 1900-1939, (3) either Historical Statistics or Coen 2000 starting in 1919, and (4) Coen 2008a ending in 2007? Also, if you wanted more recent numbers, where would you likely go? It would also be good, I think, if we could include newspaper advertising in recent years. Abernathy (2020) said that the US lost a quarter of its newspapers in the previous 15 years and half it’s newspaper reporters in the previous 10. That complements the narrative in claims in Vaidhyanathan (2018), Antisocial Media: How Facebook Disconnects Us and Undermines Democracy (Oxford U. Pr.): We’ve seen an increase in polarization in the international body politic since the elimination of the Fairness Doctrine in the US and the rise of the internet. McChesney recommends citizen-directed subsidies for journalism. Dean Baker wants to make for-profit internet companies liable for defamation for content they “boost” or otherwise promote in exchange for money. However this discussion goes, it would be useful to have data like this, preferably going back to the US Postal Service Act of 1792 if we can get it — even if it’s not great, as long as it has some plausible basis.
I plan to compile numbers that seem most relevant to these questions into a dataset to be published with the “Ecdat” package for R on the Comprehensive R Archive Network (cran.r-project.og) then post a tutorial to Wikiversity on how to update these numbers every year or so. After that, I’d be happy to draft an article that the two of us as co-authors could submit to the Real-World Economics Review; I think it would be more likely to get published and have a bigger impact if it reflected your wisdom from having worked with these numbers for many years.
Comments? Thanks again,
These are important issues. I’m working on other projects now, but I applaud your public-spiritedness in examining these issues. I don’t know of recent numbers — Facebook / Google are huge now in advertising.
Also, have you tried to communicate with “https://ourworldindata.org” about what it might take to make data on advertising part of their collection in a way that is regularly updated automatically?
Might be worth doing, but not something I have time to pursue now.