Table schema

This section describes the columns in the Ad Targeting, Ad Targeting delta, and Ad Library data tables, and provides some related clarifications.

Ad Targeting table schema

In this table, the Type (data type) can be:

• String: Sequence of characters, representing either a constant or a variable.

• Boolean: One of two possible values (usually denoted true and false).

• BigInt: Integer (numerical) value.

• List: Series of values. Angle brackets indicate the data type of the values in the list.

• Map: Series of key/value pairs. Angle brackets indicate the data type of the key and the data type of the value, separated by a comma.

[
  {
    "College grad": "Education level"
  },
  {
    "Wine": "Interests",
    "Golf": "Interests",
    "Organic food": "Interests",
    "Skiing": "Interests",
    "Security (finance)": "Interests",
    "Frequent Travelers": "Behaviors",
    "Sustainable development": "Interests",
    "Business and industry": "Interests",
    "Natural environment": "Interests",
    "Sailing": "Interests",
    "Running": "Interests",
    "SVT Play": "Interests",
    "Politics": "Interests"
  }
]

Delta table schema

The delta table lists ads that were reclassified as non-political (and therefore, removed from the table) since the last table publication. It is a single table to which we add data month to month.

Ad Targeting dataset columns for indicating missing data

As of November 1, 2021, we started tracking information about custom audiences that were deleted sometime after the ads were initially run. For data prior to that date, the Ad Targeting dataset might be missing information about whether custom audiences, datafile custom audiences, or lookalike audiences were used or not. By "missing information", we mean that one of the affected columns could read False when it should actually be True. If it already reads True, we know for sure that it is True.

We have added two columns to the table schema where we indicate whether data is missing. The following table lists the two new columns, indicates the columns of data to which they pertain, and provides the meaning of the possible values:

Location data

Throughout the Ad Targeting dataset, we provide the location targeting chosen by an advertiser for the ad. Advertisers can input location targeting in a number of ways, for example by selecting zip codes, countries, Designated Market Areas (DMAs), pin drops (longitude and latitude), addresses, and places within a specified radius. To prevent inadvertently sharing granular location of people on Facebook, we've provided the location options chosen by advertisers when these levels have been set to zip code level or greater. For smaller geographic designations, we note the type of selection (such as an address, place, or location pin drop), the city it falls in, and the radius specified by the advertiser.

The following examples help illustrate the transformations. The left column shows the targeting selection by the advertiser; the right column shows the transformation found in the data set. You will see that addresses, pin drops, and places are redacted and replaced by the text "<address>", "<location>", and "<place>", respectively.

Whenever we cannot report a specific location, we replace it with "Unknown Country", <unknown location>, or <unknown place> as in this example: {"Unknown Country":{"":{"":["<unknown place> (+5 km)"]}}}. This can happen for a number of reasons. A country, for example, could be unknown because the location is too close to a border to classify accurately, or because borders have changed. A place (such as a church or a school), could become unknown because the Facebook page from which the place's location was extracted was deleted.

Tracking ad change history

An ad's archive_id in the Ad Targeting dataset maps to the ad's ID in the Ad Library product, where researchers can access the ad creative, the Page associated with the ad, who paid for the ad ("Paid for by" disclaimer), a range of how much they spent, and the reach of the ad across multiple demographics.

If an advertiser makes changes to an ad after it has been created, (such as editing the creative, targeting logic, or spend limits), then our system stores the updated version of the ad as a new distinct ID. This means that distinct archive_id values in the dataset could relate to a similar creative, such as if the only modification was to change the targeting of the ad.

Ad Library data table schema

The Ad Targeting dataset (refreshed monthly) contains data about targeting selections for each ad, specified by `archive_id`. We also provide the Ad Library data table that contains information about ad delivery, such as what pages ran the ad, when it was created, how many users it reached, and so on. This table is refreshed daily.

To get a complete picture of the advertisement and its content, including both its targeting and its delivery, you need information from both tables. See Joining Ad Targeting dataset data with Ad Library data.

Joining Ad Targeting dataset data with Ad Library data

If you want more information about an ad (its creative, spend, and so on) and want to map the ad with its targeting information, perform a join between the archive_id column of the ad_targeting_dataset_siep_aug_2020 table (the Ad Targeting dataset) and the fbid column of the ad_archive_api table (the Ad Library dataset) as shown in this example:

data_joined = execute("SELECT * FROM fbri_prod_atp.ad_archive_api AS a, fbri_prod_atp.ad_targeting_dataset_siep_aug_2020 AS b WHERE a.ds = b.ds AND a.fbid = b.archive_id")

Inconsistencies between the Ad Targeting dataset table and the Ad Library data table

If you perform a join between the two tables, you will see that there are ads in the Ad Targeting dataset table that do not have a corresponding entry in the Ad Library data table (or vice versa). This happens because ads initially classified as political can be reclassified as non-political after they have been run.

When a reclassification occurs, the Ad Library data table reflects the change before the Ad Targeting dataset because the Ad Library data table is updated daily and the Ad Targeting dataset is updated monthly.

Beginning with the second monthly update, there is an Ad Targeting dataset delta table that describes ads that were reclassified as non-political (and therefore, removed from the dataset) since the previous monthly update.

Learn more

• Custom audience
• Customer list
• Lookalike audience
• Ad creative
• Ad Library data
• Facebook Terms of Service
• ISO currency code
• Audience distribution
• Insights range value
• Estimated audience size
• ISO 639-1 language codes