People quasi-randomly assigned to farm rice are more collectivistic than people assigned to farm wheat

The rice theory of culture argues that the high labor demands and interdependent irrigation networks of paddy rice farming makes cultures more collectivistic than wheat-farming cultures. Despite prior evidence, proving causality is difficult because people are not randomly assigned to farm rice. In this study, we take advantage of a unique time when the Chinese government quasi-randomly assigned people to farm rice or wheat in two state farms that are otherwise nearly identical. The rice farmers show less individualism, more loyalty/nepotism toward a friend over a stranger, and more relational thought style. These results rule out confounds in tests of the rice theory, such as temperature, latitude, and historical events. The differences suggest rice-wheat cultural differences can form in a single generation.


Language
The research assistant communicated with the farmers in Mandarin Chinese.The farmers commonly understand Mandarin because the local dialect is close to Mandarin.

Testing Assumptions of Statistical Tests
We tested the assumptions of statistical tests by checking for skewness, kurtosis, and equality of variance using the raw data.For the four cultural difference measures, we calculated skewness and kurtosis then compared these to cutoffs of < 2 for skewness and < 7 for kurtosis.
Relational Pairings: Skewness = -2.01,Kurtosis = 3.84 Self-Inflation Friends: Skewness = 0.75, Kurtosis = 1.97 Self-Inflation Family: Skewness = 0.74, Kurtosis = 2.23 Loyalty/Nepotism: Skewness = 0.72, Kurtosis = 2.46 All four outcomes were below the cutoffs, except that the skewness of relational pairings was 0.01 over the recommended cutoff.Because the deviation from the cutoff was so small, we decided not to transform the data.
Next, we tested for equality of variances in each outcome across rice and wheat farms using Leven's test for equality of variances in the program R. The test reports whether there is significant evidence for inequality of variances.The result of each test was non-significant.

Hui Participants
Ningxia is home to many Hui, who are a Muslim minority in China.According to the 2000 Census, Ningxia is 34% percent ethnic Hui, although only 6.5% of our sample was Hui.Compared to other Muslim populations like the Uyghurs, the Hui are quite assimilated into the majority Han culture, apart from religion.For example, most Hui speak Chinese 33 .

Chinese State Farms Relational Mobility Compared to 39 Studied Societies
Although the relational mobility scale showed low reliability, we ran tentative comparisons to other published studies of relational mobility.Because of the low reliability, these comparisons should be interpreted with caution.Figure S1 shows the basic differences in relational mobility between the rice and wheat farms.Next, we compared our relational mobility results to a recent study of 39 societies around the world 21 .We did this because we know of no prior study that has tested relational mobility in farming communities.Researchers have argued that farming is a low relational mobility subsistence style 21 .However, previous studies have relied on college samples or samples from Facebook 21 .Thus, our farmer data offers a rare opportunity to compare relational mobility in a farming community with the student and online samples common in psychology.
The farmers in China reported lower relational mobility (M = 3.84, SD = 0.57) than every other society tested (Figure S2).Note that this is a descriptive statistic.It is not a formal statistical test comparing this new data to the previous study in 39 cultures 21 .We suggest that future studies can check the robustness of this finding with other samples.

PEOPLE ASSIGNED TO FARM RICE MORE COLLECTIVISTIC
If farmers have lower relational mobility than all previous cultures studied, it would suggest two interesting implications for theory: 1. First, it would be consistent with the theory that farming is a low relational mobility subsistence style 21 .
2. Second, it would be consistent with the idea that online samples common in psychology are giving us a narrowed view of the range of data across societies.This would suggest that widening our samples to include more diverse populations, such as this farming sample, would widen our view of human psychology.This is echoed by two recent papers advocating for sample diversity in psychology.For example, one researcher argued recently that we should consider sample diversity beyond the WEIRD vs. non-WEIRD dichotomy 34 .For example, China is a non-WEIRD culture.But the difference in holistic thought between this farm sample and our previous university samples in China is nearly as large as the difference between our university samples in the US and China.This speaks to the benefit of getting samples not just outside the West, but with hard-to-reach populations.

Statistical Power
We recruited as many farmers as we could approach given the limitations of our field site and recruitment.To give readers a sense of statistical power, we conducted a sensitivity analysis.With the final sample size of 234 participants, 80% statistical power, and two-tailed p values, the sample would be able to reliably detect effects as small as r = 0.18.

Holistic Thought Among Farmers in China Compared Tea Farmers in Turkey
Our holistic thought data can be compared to tea farmers in Turkey from a study that used the same task 18 .The Turkish study is the only other study we know of that tested farmers on this task.The farmers in Turkey from the previous study chose 68% holistic pairings 18 .Averaged across the rice and wheat farms, our participants averaged 87% holistic pairings.Note that this is a descriptive comparison.We cannot run a statistical comparison because the original study did not report the original data or enough statistical information for us to run a statistical test of these two samples.
Comparing different farming populations may be useful for subsistence theory.

PEOPLE ASSIGNED TO FARM RICE MORE COLLECTIVISTIC
The intensiveness of farming may be an important difference between farming systems.both rice and wheat farming in China are more intensive than wheat farming in Europe 2 and tea farming in Turkey.Thus, if this theory is correct, the more intensive forms of farming in China should produce more holistic thought than less intensive forms of farming.Of course, there are other differences between the farming communities in Turkey and China besides the intensity of the crops.Thus, we would need more research to test whether intensiveness is important.

Figure S3
Leftover Corn Straw on the Qukou State Farm Note: The Qukou State Farm grows mainly dryland crops like wheat and corn (pictured), which require less labor than paddy rice and do not depend on irrigation.

Livestock
Studies have found that herding cultures tend to be more individualistic than farming cultures 18,35 .In the two state farms, animal husbandry makes up only a small portion of the agricultural economy (12% in Lianhu and 8% in Qukou according to statistics we could find from 1988).Furthermore, this is livestock raising, rather than mobile herding.Livestock raising may be closer to farming than the mobile herding of cultures like the Mongolians and Tibetans.

Farm Samples and Backgrounds
We recruited farmers by working through division leaders on the farms in summer and fall of 2017.Each farmer belongs to a division, which has a division leader who oversees administrative tasks.Our team explained the study to the division leaders, and the leaders passed along the information to the farmers in their division.
Farmers who were interested in participating contacted our team, and we set up a testing session.
Nearly all of the participants (96.2%) were currently farming (which we defined PEOPLE ASSIGNED TO FARM RICE MORE COLLECTIVISTIC as having farmed land in the year of the study).The other nine participants were helping with administrative tasks for the farm.
Most participants (73.5%) grew up on the farms (Lianhu or Qukou).That means that most of the participants were the descendants of farmers assigned by the government to the farms.That also means that most participants were lifelong farmers.
However, about one-quarter (26.5%) were assigned to the farms after growing up elsewhere.
In this sample, the percentage of local born was slightly higher on the wheat farm (76.9%) than the rice farm (69.2%).This could make it seem like there are differences between the farms in the percentage of newcomers.However, we later returned to the farms to collect data for a separate project, and that sample showed the reverse pattern-more local-born people on the rice farm (79.8%) than the wheat farm (73.1%).Adding the two samples together reveals a nearly identical percentage of local-born farmers on the rice farm (74.5%) and wheat farm (75.2%).This is consistent with the data on population size of the two farms showing that the populations were mostly stable after 1980 (Supplementary Notes Section 9).
We ran regressions testing whether rice-wheat differences were larger or smaller for participants who were born on the farms.There are logical reasons to expect that differences could be in either direction.On the one hand, people born on the farms might be more enculturated because they are second generation.On the other hand, there is some evidence that cultural differences grow larger as people age 36 , even into adulthood 37 .Thus, the original people assigned to the farms may have had more time to enculturate.
Regardless, among the tasks that showed significant rice-wheat differences in Table 3, we found no significant interactions that would support the idea that rice-wheat differences depended on whether people were born on the farms (tested as an interaction between rice farm and a dummy variable for born locally).Some of the means in the data were in opposite directions.For example, there was a non-significant interaction for self-inflation on the family sociogram in the direction that rice-wheat differences were larger among people not born on the farm (B = 3.04, t[187] = 1.94,P = 0.054, 95% CI [-0.05, 6.13]).However, loyalty-nepotism differences showed a non-significant trend in the opposite direction-slightly larger among people born on the farm (B = 1.05, t[187] = 0.70, P = 0.486, 95% CI [-1.92,PEOPLE ASSIGNED TO FARM RICE MORE COLLECTIVISTIC

4.02]).
In sum, we found no significant interactions indicating that rice-wheat differences were larger or smaller for people born on the farm.The next section takes a different angle to this question: are rice-wheat differences larger or smaller among younger farmers?Analyzing age led to a similar conclusion.
One small note about the sample location: Ningxia is technically an autonomous region.However, for practical purposes, people in China think of it the same as other provinces.The same is true of other autonomous regions, like Xinjiang, Tibet, and Inner Mongolia.

No Evidence That Rice-Wheat Differences Are Weakening Over Time
The main text reports that the relational mobility questionnaire did not reach acceptable reliability.We suspected this could be because the questions are somewhat complex and difficult to understand, especially in an environment where people are not used to taking questionnaires.However, as living standards have risen in China, the younger farmers may have more exposure to writing, which could make the questionnaire easier for them to complete.
In line with this idea, we found that the scale reliability improved from 0.45 in the entire sample to 0.55 among farmers below 47 years old (the average age of the sample).In addition, rice-wheat differences were larger among younger participants than older participants (age x rice interaction in a regression: B = 0.04, t[184] = 2.69, P = 0.008, r = 0.20, 95% CI [0.01, 0.06]).This fits with the idea that comprehension was a problem for farmers with the relational mobility questionnaire.
We did not find significant interactions of age with rice for implicit individualism, holistic thought, or loyalty/nepotism.It is interesting that these results do not fit with the idea that rice-wheat differences are disappearing over time as China modernizes and the market economy reaches into more parts of Chinese society.Instead, rice-wheat differences were just as large among younger farmers.

Demographic Differences Between Rice and Wheat Farm Samples Before and After Propensity Score Matching
We used the MatchIt package in the program R (version 4.5.1) to do propensity score matching with optimal matching.This improves on the more basic method of nearest neighbor matching.Simple nearest neighbor matching finds the optimal match for the first participant, regardless of whether that leaves less-optimal matches for subsequent participants.In contrast, optimal matching finds matches while trying to make the entire sample more optimally matched.
We ran t tests to compare demographic differences before and after matching the samples from the rice and wheat farms.We tested for differences in participants' self-reported gender identity, age, family income, and maternal educational attainment.Note: This table tests for demographic differences between the samples from the rice farm and wheat farm before using propensity score matching (left) and after (right).
Negative t values mean the rice farm scored lower than the wheat farm; positive values mean the rice farm scored higher.These tests are for the matching with the participants who completed the triad task.For the other tasks, the numbers differ slightly because of missing data.Analyses are independent samples t tests with two-tailed p values, without correction for multiple comparisons.

Farm Populations Over Time
In this section, we detail the history of the founding of the farms and their populations over time in more detail.This increased the official population of the farm, but this area is separate from the main farm.When recruiting participants, we did not work with any team leaders in the Taiyang Liang area.

No Statistically Significant Differences Based on Test Setting
We tested most farmers (93.2%) in office rooms on the state farms.At the beginning of the study, we tested farmers at home, but we soon learned that the home environment was distracting.For example, we could not prevent other family members from entering the home during the tests.Therefore, we started testing the farmers in office rooms on the farms, where we could get a more quiet environment.
A research assistant was present regardless of setting.
We analyzed the data to test whether the setting influenced the results.Table S2 reports regression analyses of whether the setting of the test (home or in an office) was related to the results.We used a dummy variable where 1 = the participant completed the tests at home and 0 = at an office.
The results show that test setting was not statistically significantly related to any of the outcomes.Differences in holistic thought style between people tested at home or in the office were not statistically significant (B = 0.32, t[180] = 1.27,P = 0.203, 95% CI [-0.15, 0.85]).Rice-wheat differences remained significant after controlling for test setting.

Figure S1 :
Figure S1: Relational Mobility on the Rice and Wheat Farms Figure S2: Relational Mobility in the Farm Samples Compared to 39 Societies Figure S3: Leftover Corn Straw on the Qukou State Farm

Figure S1 Relational
Figure S1Relational Mobility on the Rice and Wheat Farms

Figure S2 Relational
Figure S2Relational Mobility in the Farm Samples Compared to 39 Societies

Table S1 :
Demographic Differences Between Rice and Wheat Farm Samples Before and After Propensity Score Matching

Table S2 :
Rice-Wheat Differences Remain After Controlling for Test Setting

Table S3 :
Rice-Wheat Differences Without Control Variables or Matching (Zero

Table S4 :
Descriptive Statistics PEOPLE ASSIGNED TO FARM RICE MORE COLLECTIVISTIC

Table S5 :
Descriptive Statistics for Binary Variables

Table S6 :
Farmers' Own Educational Attainment Is Not Significantly Related to

Table S1 Demographic
Differences Between Rice and Wheat Farm Samples Before and After We were able to locate yearly population records for the Lianhu rice farm through 2008.The farm population grew from its founding until reaching 5,960 people by the end of the Cultural Revolution in 1977.The statistics show that the population of the farm was mostly stable from 1977 on, hovering around 6,000 people.The last data point shows 5,946 people in 2008.Although this data does not rule out the possibility of people moving in and out, it suggests that the farm population has been largely stable for the last 40 years.Qukou added a parcel of land called Taiyang Liang 15 kilometers from the main farm.The government used this land to settle farmers whose land in the Guyuan area of Ningxia suffered from desertification and ecological degradation.
This table tests whether rice-wheat differences remained after controlling for whether participants took the tests in an office (0) or at home (1).Samples are matched using propensity score matching.Female, Hui, and Rice Farm are coded 0 = no, 1 = yes.Age is in years.Hui are a Muslim religious group in China.Degrees of freedom = 188, except 189 for the friend sociogram and 187 for holistic thought.Analyses are regressions with two-tailed p values. Note: