Which Country Do I Look Like? A Question About Comparison
A face does not carry a country inside it; the sentence only works once you name the comparison set, and that set changes the verdict. Psychologists have measured the asymmetry since 1977.

The sentence has a direction
Ask a stranger which country they look like and you have already chosen the direction of the comparison. Psychologists treat that choice as consequential. Amos Tversky's 1977 paper in Psychological Review argued that similarity is not a distance but a match between features, and one prediction of the model is that the relation is asymmetric. In one of his studies, 66 of 69 people agreed that one country of a pair resembled the other, while only 3 would say it the other way round. The direction people preferred ran from the less prominent member of the pair toward the more prominent one. Tversky described the asymmetry as a general property: the variant resembles the prototype more than the prototype resembles the variant.
That asymmetry has a plain consequence for the question in the title. Asking which country you look like and asking which country looks like you are not the same sentence, and the evidence suggests they do not get the same answer. The first phrasing treats the face as the variant and the country as the reference; the second reverses the roles. Tversky's finding was that respondents systematically favour one of those directions, and which one depends on the relative salience of the two things being compared. A question phrased as a request for a country is therefore not a neutral measurement. It is a request to place a face against a standard that the sentence has already elevated.
Which features count depends on the list
Similarity also moves when the alternatives change. Tversky's paper reported a sorting experiment built from human faces, in which a target face was judged against a small set that varied by one member. A feature shared by everything in a set cannot help distinguish anything inside it; his own illustration is the word real, which separates nothing among actual animals but becomes informative the moment a centaur or a phoenix joins the group. The measured effect was modest but reliable, averaging around nine percent on the sorting measures across the sets he tested.
Translated into the question at hand, the classification depends on the candidate list. Tversky showed that people sorted a pair of countries together at strikingly different rates depending on which other countries were present: in one condition a pairing was chosen by 60 percent of respondents, and in another the same decision was reached by 96 percent of them once a single alternative had been swapped out. The face has not changed. The list has. Any tool that returns a single country name from an image has therefore made a choice that the question did not specify, and the honest way to state that choice is to publish the list alongside the answer.
Two differences feel like more
A related result removes the comfortable assumption that resemblance adds up. In a 1982 paper in the same journal, Tversky and Itamar Gati tested stimuli that varied along two separate dimensions and found that objects differing on both were judged more dissimilar than the sum of the two single differences predicted. They called it the coincidence effect and demonstrated it across house plants, parallelograms, histograms, and schematic faces. The practical reading is that similarity is not a tidy arithmetic of features. When a face differs from a reference on two independent grounds at once, the impression of difference is amplified rather than merely added, which is one reason a face can feel decisively unlike a region and still share most of that region's measured traits.
This is not a quirk of laboratory stimuli. It is the ordinary behaviour of a comparison made without a specified yardstick. Two faces that differ on nose width and eye aperture are read as more distinct than either gap alone would justify, because the mind treats the co-occurrence as information. The same amplification works in reverse in a quiz: a face that matches a reference on two salient features feels like a strong hit, and the player stops looking for the features that disagree. Neither the amplification nor the confidence is a mistake in perception. Both are the expected output of a system that has been asked to compare without being told which dimension matters.
A country is a graded category
Country names are categories, and categories are not uniform. Eleanor Rosch showed in the 1970s that membership is graded rather than all-or-nothing: some members of a category are better examples than others, and people agree closely on which. Her 1973 paper in Cognitive Psychology and a 1975 paper in the Journal of Experimental Psychology: General established that reaction times, list order, and priming all track how typical an example is. The philosophical groundwork was older. Wittgenstein had argued in 1953 that many ordinary categories hold together by criss-crossing resemblance rather than by a shared essence, and Rosch's experiments gave that argument a measurement.
That matters because a country is a poor category for a quiz answer. National borders were drawn recently and provisionally, while the catalogue's unit is a described pattern anchored to physical geography, so the two vocabularies do not nest. Senegal's name appears inside the record for Senegalid, which describes a gracile Sudanid variety named by Lundman in 1952 and recorded across the Sahel rather than inside one border. West Amazonid is named for a rainforest. Moorish is named for a people, and its own record adds that the type is not stabilised and carries variable features. Asking which country a face resembles presupposes that the answer set is countries, when the reference set is landforms, rivers, and regional belts.
The reference is the average
There is one more structural problem, and it belongs to face perception rather than to categorisation. In 1991 Tim Valentine proposed in the Quarterly Journal of Experimental Psychology that faces are encoded as points in a multidimensional space, and that a face's distinctiveness is essentially its distance from the centre of that space. Faces near the middle are typical, densely clustered, and easy to confuse with one another; faces further out are distinctive and are recognised faster. His framework offered two variants with the same predictions, one storing a norm or prototype, the other storing only exemplars, and he noted that the experiments he ran could not separate them.
The consequence for a country quiz is uncomfortable. A country composite is built by averaging many portraits, which places it close to the centre of whatever sample was averaged. The reference is therefore, by construction, the most typical face the sample produced, and typical faces are the ones observers find hardest to tell apart. Asking whether a face looks like a country average is asking whether a face resembles the least distinctive point available. The comparison still functions, because people distinguish coarse regional tendencies well above chance, but the ceiling is set by the method rather than by the player's attention.
The catalogue is a dated reference set
This site's own answer to the question is a set of 209 described patterns rather than a set of countries. Each entry carries a region, a list of sourced traits, and the printed works it was drawn from, and each is anchored to roughly 1500 years ago. The entries also include recent contact types, which is worth stating plainly: Arizonid is recorded as having developed as Athabaskan speakers moved south in the late pre-Columbian period, and as becoming more numerous once the horse returned to the continent; its record calls it relatively recent and still numerous. The reference set is therefore not a snapshot of deep antiquity. It is a layered document whose newest layers are only a few centuries old.
What the reference set cannot do is follow people. Borders moved, trade moved, and modern populations are mixtures at every scale, so a pattern documented in one belt may now describe a minority of the people living there while describing parts of a city on another continent. The catalogue says as much in its own framing: a pattern is a statistical tendency in a historical sample, not a claim about a living population, a nationality, or an individual. That is why a resemblance answer is best read as a statement about a document. Which pattern does this face most resemble is answerable with a range; which country is this person from is not answerable from a photograph at all.
Frequently Asked Questions
- Does a face really look like a country?
- Not in the way the sentence implies. A country is an administrative area, and a face is a bundle of features whose frequencies vary gradually across geography. When a tool or a person names a country, they are comparing the face to a reference sample and reporting which regional average it sits closest to. That is a resemblance claim about a stored reference, not a fact about the person, and it can change when a different reference set is used.
- Why do two different quizzes give two different countries?
- Because each one carries its own reference set, and similarity depends on the alternatives available. Tversky demonstrated in 1977 that people's similarity judgments shift when one item in the comparison set is replaced, with an average effect of roughly nine percent in his sorting studies. A quiz built on national averages, one built on continental composites, and one built on historical patterns are answering different questions with different yardsticks. Disagreement between them is expected, not evidence of an error.
- Is there a better way to phrase the question?
- Yes, and it is mostly a matter of naming two things instead of one. Instead of asking which country a face looks like, ask which documented regional patterns it most resembles and how closely, and state which reference set you are using. That version can be answered with a range, it can be revised when a better sample appears, and it does not require the answer to correspond to a state. It also makes the comparison falsifiable, which is the property that turns an impression into a claim.
Related Phenotypes
Faces from the encyclopedia that appear in this article. Open any entry for its full description, distribution, and references.
Arizonid
North America
Relatively recent North American contact type. Started to develop as Athabaskan speakers of Canada migrated South during late Pre-Columbian ...
Senegalid
West Africa
Gracile Sudanid variety, named after the country of Senegal, where it is found in its purest form among Wolof, more mixed in Serer. Occasion...
West Amazonid
South America
Amazonid subtype, the most frequent type in the Western Amazon rainforest. Common among Arawakan speakers, typical groups are Yawalapiti, As...
Moorish
North Africa
Relatively recent contact type of the Western Sahara. Probably formed during Antiquity when aboriginal Berber populations mixed with older W...
Test what you learned
Put your eye for regional faces to work in the daily quiz — a composite face, a world map, and your best guess.