The analysis in this README was written by an AI (Claude, working as Rapidata's engineering assistant), and every number was computed by a script from the data in this repo. The data itself is real: 10,000 human responses collected through Rapidata. Treat the interpretation as exploratory, not peer-reviewed.

We asked 10,000 people in 127 countries one question:

Would you rather never kiss someone on the mouth again, or never eat anything with sauce again?

Result: 53.1% would give up kissing to keep sauce (95% CI ± 1.0 pp). Globally it is close to a coin flip, but the answer depends a lot on where people live: it ranges from 35% in PT to 71% in EG.

- Country and culture explain far more than anything else (χ² p = 1.9e-55). The Arab world (68%) would give up kissing; Iberia (38%) would give up sauce.

- This isn't caused by translation. In Arab countries, people who saw the task in English or French still chose "give up kissing" far more often than the global average.

- Women give up kissing more often than men: after controlling for country, men have an odds ratio of 0.73 vs women (p = 4e-07).

- Age has no real effect. The raw dip among 50–64 year olds comes from who is in that age group: 44% of them are European, vs 15% of 18–29 year olds. After controlling for country, no age group differs significantly.

- Reliability score doesn't change the answer: weighted by user_scorethe result is 53.1%, vs 53.1% unweighted.

One row per response. No user identifiers are included.

from datasets import load_dataset

ds = load_dataset("jasoncorkill/would-you-rather-kissing-vs-sauce", split="train")

df = ds.to_pandas()

df.groupby("country").gives_up_kissing.mean().sort_values()

A CSV copy is at data/responses.csv.

Every percentage below is the share who would give up kissing (and keep sauce). Confidence intervals use the normal approximation. Groups with fewer than 50 respondents are left out of the tables.

χ² test of independence: p = 1.8e-67.

χ² p = 1.9e-55. The spread between EG and PT is 36 percentage points, far larger than either confidence interval.

χ² p = 1.3e-54. The language ranking closely follows the country ranking. Arabic is at the top, and Spanish and Portuguese are at the bottom.

If the Arabic translation were driving the effect, people in Arab countries who saw the task in another language would look like the global average. They don't:

Arabic-language respondents lean further toward giving up kissing (χ² p = 2.1e-05), but English- and French-language respondents in the same countries still sit well above the global 53.1%. The translation may add to the effect, but it doesn't create it.

χ² p = 4e-06. A logistic regression that controls for country (the top 15 countries plus "other") gives men an odds ratio of 0.73 vs women (p = 4e-07), so the gap isn't explained by gender mix across countries.

In most countries women are more willing to give up kissing than men. Where a country breaks that pattern, the table shows it.

The raw χ² is p = 0.005, but this effect largely comes from who is in each age group. Older respondents are far more often European:

After controlling for country, no age group differs significantly from 18–29:

Within each region the share is mostly flat across ages. The one outlier is Other Europe 65+ (55%, n = 168); that is a small cell, so treat it with caution:

χ² p = 0.026. The occupation list mixes jobs, education level and employment status, and the differences are small next to the country effect. Read it as descriptive only.

user_score quintiles:

χ² p = 0.337. There is no meaningful trend: careful and less careful respondents answer the same way.

- The sample is not representative. Respondents are people using apps in Rapidata's network, not a probability sample. IN, ES, PH, EG alone make up 46% of responses. The global 53.1% is not a population-weighted world estimate. Several large countries have very few respondents: US 57, CN 11, ID 25, NG 7, MX 45, RU 0.

- Many respondents have no demographic data: 49% have no age and 38% have no gender. The age and gender analyses cover only those who reported them, who may differ from those who didn't.

- Machine translation: the wording is not identical across languages ("sauce" covers different foods in different cuisines, and "kiss on the mouth" carries different cultural weight).

- This is a hypothetical question. People answered quickly in an app, so treat the results as a gut reaction.

- No multiple-comparison correction was applied across the many breakdowns; focus on the large effects (country, region, gender) rather than individual small cells.

@misc{rapidata_kiss_vs_sauce_2026,

title = {Would You Rather: Kissing vs Sauce},

author = {jasoncorkill},

year = {2026},

url = {https://huggingface.co/datasets/jasoncorkill/would-you-rather-kissing-vs-sauce}

}

Responses collected with Rapidata.

- Downloads last month

- 113