In 2014, Martin Gilens and Benjamin Page published a highly influential paper which looked at 1,779 policy issues in the United States from 1981 to 2002. In it, the authors examined how much of an effect the independence preferences of average citizens (50th percentile), affluent citizens (90th percentile), and interest groups had on policy change. The result? Average citizens were basically irrelevant, interest groups seemed to matter at the thresholds of net supporting or opposing a policy, and affluent Americans had the vast majority of the power.
In 2015, Martin Gilens did a follow-up study which expanded the number of policies to 2,245, the time span from 1964 to 2006, and included a separate result for poor Americans (10th percentile). Once again, affluent Americans were dominant in influence while average and poor Americans had little influence.
At the time of these findings, the media was in a state of frenzy over it. It received tons of attention and coverage, but then got strangely memory-holed. Unsurprisingly, because of how controversial these findings were, various criticisms have been made about them. One of them comes from Branham et al. (2017), who noted that in situations where those with middle and high income disagreed, middle income Americans won 47% of the time. However, this is misleading. As Bowman (2020) explains, win rates suffer from several issues, which I will list below:
Status quo bias, which, as its name implies, is the common finding that the United States government has a preference to change nothing, even when change is preferred by its own populace. This was one of the responses made by Gilens, though his critics argue that this alone didn’t seem to alter their results very much.
Defining thresholds, which was done differently between Gilens and his critics, and different definitions result in substantially different findings.
Preference gaps and the cutoff used also affects the results. Does a 5% difference count? 10%? 25%?
When a policy change does get occur even with disagreements between average and affluent Americans, that alone does not show that this was a “win” on the part of the group which supported the change. This is because there is always a baseline probability in which a policy change gets adopted even when neither average nor affluent Americans support it.
Bowman considers all of this in his paper, and what does he find? The trend for average Americans on policy change was extremely similar to if neither average nor affluent Americans preferred it, implying their effect was basically zero. By contrast, the trend for affluent Americans on policy change was extremely similar to if both average and affluent Americans preferred it, implying they had almost all the say.
In Bowman’s own words:
Averaging across all 756 cases of disagreement, policies supported by the middle alone are enacted at roughly the same rate (24.5 percent) as those opposed by both groups, and the policy change rate for those policies supported by the affluent alone (37.5 percent) is about the same as for those supported by both groups. This overall finding holds even when using the majoritarian definition of policy disagreement that Gilens’s critics use (without a specified preference gap). The policy change rate using this definition is significantly higher when the rich support change (36.5 percent) than when neither group does (22.9 percent), but not when the middle support change (25.6 percent). Conversely, the difference in policy change rates between cases where both groups support change (39.4 percent) and cases where one group opposes change are significant when it is the rich opposing change (−13.8pp) but not when the middle oppose change (−2pp). These findings offer strong evidence for Gilens’s claim that while high-income preferences have a sizable independent impact on policy, the preferences of the middle have little to no impact.
So, after all this back and forth, the best evidence we have suggests that Gilens was correct all along.
What about outside of the United States? Europe is relatively more progressive, so surely the average citizens would matter more there, right? Apparently not. Schakel (2021) examined policy responsiveness in the Netherlands, and he basically replicated Gilens’ findings:
Predicted probability of policy change by income when preferences diverge (dotted lines indicate 95% confidence intervals).
This finding is not driven by differences in likeliness to vote, as Shackel noted:
For voters at the 50th income percentile, for example, an increase in support of 1% multiplies the odds of a policy change being enacted by about 1.026 (P = 0.001). For non-voters at the 50th percentile, this is only 1.014 (P = 0.032). However, when estimating a model with the preferences of low-income voters and high-income non-voters as independent variables, the effect for high-income non-voters stays intact (OR = 1.018, P = 0.038), whereas the effect for low-income voters disappears (OR = 1.003, P = 0.759).17 This suggests that controlling for voting at national elections does not eliminate the income bias in responsiveness. In fact, it is more important to have a high income than to vote when one wants their views to be implemented.
Lupu & Castro (2022) explored the results in Spain using educational and occupational differences as their comparisons, and sure enough, similar findings. As a matter of fact, their findings are arguably more shocking because for the lowest performing group in both categories, their effects were negative. In other words, if poor citizens wanted a policy adoption, that policy is actually less likely to get adopted.
Here were the effects on policy changes broken down by policy type:
So, the effect elitism seems to be the strongest on in Spain appears to be on the economy and cultural issues. For political processes and security and foreign affairs, however, it doesn’t matter very much. Notice the wide confidence intervals. This is due to a small number of observations available for each analysis. As a result, none of the policy types except for cultural issues reaches the conventional threshold for statistical significance. If I were to bet money on it though, elites probably have a real positive effect on economic issues as well, as this is in line with previous research.
Mathisen et al. (2023) was able to replicate the findings of Gilens to Germany, the Netherlands, Norway, and Sweden. This should be surprising considering the relatively more egalitarian nature of Scandinavian countries such as Norway and Sweden, but nonetheless, substantial unequal policy responsiveness was still observed. When looking at the instances where the preferences diverged between the 90th percentile, 50th percentile, and 10th percentile, across all the European countries, the 90th percentile was favored over both the 50th and 10th percentile.
What is interesting to note is that in the appendix of this paper, in the case of Germany and Sweden, the net influence of those at the 10th percentile seems to be negative. Norway is overall the best egalitarian in terms of responsiveness compared to the other European countries, but it is still a far cry from the idealized notion of a government that cares about everyone equally, as the authors themselves noted that responsiveness to the preference of the 90th percentile is 2.5 times greater than the 50th percentile.
Lastly, the authors also find that prior to 1998, left-wing governments were, unsurprisingly, relatively more responsiveness to the preferences of poor and middle income citizens when it comes to economic and welfare policies, but this is no longer the case since then.
What about good ol’ Switzerland? The country which has often been praised for being a bastion of direct democracy, surely, they are also an exception here, right? Turns out, not entirely. As Wagner (2021) found, in Switzerland, a similar finding appears as the rest of the West: once you account for the fact that preferences between various groups are correlated by controlling for the preferences of each group, not only does the effect of the 90th percentile remain, it seemingly gets stronger, while for the 10th and 50th percentiles, their effect disappears. A similar pattern is seen when it comes to comparing different education levels in Switzerland:
However, one caveat to Switzerland is that when it comes to policy issues addressed by popular initiatives, then the results are much more egalitarian, and in the case of income, it would appear that the 90th percentile actually have the least influence. So, Switzerland does seem to have something going for it, though it seems to only apply when it comes to popular initiatives.
Regardless of what your opinion is on the notion of our modern idealization of liberal democracy, the reality is that for the most part, it does not really exist aside from in our own imagination. To the extent that the average person gets anything they want, this is mostly just a coincidence driven by the fact that oftentimes, the average citizen and the affluent citizen agree on a policy issue. Even though Switzerland is a standout case because popular initiatives seem to give the average citizens a voice, at the end of the day, this only matters if you think equal representation is a good thing. Elitism isn’t inherently bad, and in most democracies, voting, the proud tradition that defines democratic societies, is already somewhat selected. In the United States, non-voters have lower income, lower education levels, are younger, have less trust in the system, and are less likely to identify with either of the two main parties (Pew Research Center, 2014; Thomson-DeVeaux et al., 2020). There is evidence to suggest that people who are more politically towards the center are less likely to be politically literate compared to radicals (Pew Research Center, 2021). A similar case can be made for Swiss democracy. It is exceptional for one simple reason: low voter turnout, which holds true even after accounting for its large population of foreign nationals:
U.S. voting-age population turnout is still behind many other countries despite its recent rise, though registered-voter turnout is remarkably higher
Pew Research Center (2022)
Who votes and who doesn’t on the national level in Switzerland? Well, Blais (2014) found that compared to Germans, the Swiss are less likely to vote if they perceive government and politics as complicated, indicating clear evidence of selection.
We also have evidence that granting women suffrage resulted in increased welfare spending throughout Western countries (Kirkegaard, 2019). This should not be surprising at all, as democracy represents the lowest common denominator of all its constituents, so when the composition of its constituents change, we can predict what the effect will be based on the changes in constituency. The truth is simply that democracy is not some wonderful magical system. We should not expect perfectly equal representation between various groups in society, nor is that a desirable outcome we should be aiming to achieve. Elitism is the norm and will likely remain that way for the foreseeable future. The real problem is not that elitism exists, it’s the fact that the elites are delusional and out of touch with reality. People are only starting to feel discontent with the state of “democracy” because the disconnect between elite and average interests are growing increasingly apparent. In Charles Murray’s book Coming Apart: The State of White America, he presents evidence of a divergence which has occurred since the 1960s among classes. Affluent Americans have grown ever less likely to have interacted with Americans of lower classes, instead living in their own gated communities and ignorant of the real world around them. All the while low income Americans see their situation deteriorating as time passes by, the elites become ever more out of touch with the rest of America. It would be interesting to see where the future leads to once the elites and the average citizens have grown so far apart in their preferences, that even coincidental “victories” on policy issues for the latter become a rare occurrence.
In 2015, Martin Gilens did a follow-up study which expanded the number of policies to 2,245, the time span from 1964 to 2006, and included a separate result for poor Americans (10th percentile). Once again, affluent Americans were dominant in influence while average and poor Americans had little influence.
At the time of these findings, the media was in a state of frenzy over it. It received tons of attention and coverage, but then got strangely memory-holed. Unsurprisingly, because of how controversial these findings were, various criticisms have been made about them. One of them comes from Branham et al. (2017), who noted that in situations where those with middle and high income disagreed, middle income Americans won 47% of the time. However, this is misleading. As Bowman (2020) explains, win rates suffer from several issues, which I will list below:
Status quo bias, which, as its name implies, is the common finding that the United States government has a preference to change nothing, even when change is preferred by its own populace. This was one of the responses made by Gilens, though his critics argue that this alone didn’t seem to alter their results very much.
Defining thresholds, which was done differently between Gilens and his critics, and different definitions result in substantially different findings.
Preference gaps and the cutoff used also affects the results. Does a 5% difference count? 10%? 25%?
When a policy change does get occur even with disagreements between average and affluent Americans, that alone does not show that this was a “win” on the part of the group which supported the change. This is because there is always a baseline probability in which a policy change gets adopted even when neither average nor affluent Americans support it.
Bowman considers all of this in his paper, and what does he find? The trend for average Americans on policy change was extremely similar to if neither average nor affluent Americans preferred it, implying their effect was basically zero. By contrast, the trend for affluent Americans on policy change was extremely similar to if both average and affluent Americans preferred it, implying they had almost all the say.
In Bowman’s own words:
Averaging across all 756 cases of disagreement, policies supported by the middle alone are enacted at roughly the same rate (24.5 percent) as those opposed by both groups, and the policy change rate for those policies supported by the affluent alone (37.5 percent) is about the same as for those supported by both groups. This overall finding holds even when using the majoritarian definition of policy disagreement that Gilens’s critics use (without a specified preference gap). The policy change rate using this definition is significantly higher when the rich support change (36.5 percent) than when neither group does (22.9 percent), but not when the middle support change (25.6 percent). Conversely, the difference in policy change rates between cases where both groups support change (39.4 percent) and cases where one group opposes change are significant when it is the rich opposing change (−13.8pp) but not when the middle oppose change (−2pp). These findings offer strong evidence for Gilens’s claim that while high-income preferences have a sizable independent impact on policy, the preferences of the middle have little to no impact.
So, after all this back and forth, the best evidence we have suggests that Gilens was correct all along.
What about outside of the United States? Europe is relatively more progressive, so surely the average citizens would matter more there, right? Apparently not. Schakel (2021) examined policy responsiveness in the Netherlands, and he basically replicated Gilens’ findings:
Predicted probability of policy change by income when preferences diverge (dotted lines indicate 95% confidence intervals).
This finding is not driven by differences in likeliness to vote, as Shackel noted:
For voters at the 50th income percentile, for example, an increase in support of 1% multiplies the odds of a policy change being enacted by about 1.026 (P = 0.001). For non-voters at the 50th percentile, this is only 1.014 (P = 0.032). However, when estimating a model with the preferences of low-income voters and high-income non-voters as independent variables, the effect for high-income non-voters stays intact (OR = 1.018, P = 0.038), whereas the effect for low-income voters disappears (OR = 1.003, P = 0.759).17 This suggests that controlling for voting at national elections does not eliminate the income bias in responsiveness. In fact, it is more important to have a high income than to vote when one wants their views to be implemented.
Lupu & Castro (2022) explored the results in Spain using educational and occupational differences as their comparisons, and sure enough, similar findings. As a matter of fact, their findings are arguably more shocking because for the lowest performing group in both categories, their effects were negative. In other words, if poor citizens wanted a policy adoption, that policy is actually less likely to get adopted.
Here were the effects on policy changes broken down by policy type:
So, the effect elitism seems to be the strongest on in Spain appears to be on the economy and cultural issues. For political processes and security and foreign affairs, however, it doesn’t matter very much. Notice the wide confidence intervals. This is due to a small number of observations available for each analysis. As a result, none of the policy types except for cultural issues reaches the conventional threshold for statistical significance. If I were to bet money on it though, elites probably have a real positive effect on economic issues as well, as this is in line with previous research.
Mathisen et al. (2023) was able to replicate the findings of Gilens to Germany, the Netherlands, Norway, and Sweden. This should be surprising considering the relatively more egalitarian nature of Scandinavian countries such as Norway and Sweden, but nonetheless, substantial unequal policy responsiveness was still observed. When looking at the instances where the preferences diverged between the 90th percentile, 50th percentile, and 10th percentile, across all the European countries, the 90th percentile was favored over both the 50th and 10th percentile.
What is interesting to note is that in the appendix of this paper, in the case of Germany and Sweden, the net influence of those at the 10th percentile seems to be negative. Norway is overall the best egalitarian in terms of responsiveness compared to the other European countries, but it is still a far cry from the idealized notion of a government that cares about everyone equally, as the authors themselves noted that responsiveness to the preference of the 90th percentile is 2.5 times greater than the 50th percentile.
Lastly, the authors also find that prior to 1998, left-wing governments were, unsurprisingly, relatively more responsiveness to the preferences of poor and middle income citizens when it comes to economic and welfare policies, but this is no longer the case since then.
What about good ol’ Switzerland? The country which has often been praised for being a bastion of direct democracy, surely, they are also an exception here, right? Turns out, not entirely. As Wagner (2021) found, in Switzerland, a similar finding appears as the rest of the West: once you account for the fact that preferences between various groups are correlated by controlling for the preferences of each group, not only does the effect of the 90th percentile remain, it seemingly gets stronger, while for the 10th and 50th percentiles, their effect disappears. A similar pattern is seen when it comes to comparing different education levels in Switzerland:
However, one caveat to Switzerland is that when it comes to policy issues addressed by popular initiatives, then the results are much more egalitarian, and in the case of income, it would appear that the 90th percentile actually have the least influence. So, Switzerland does seem to have something going for it, though it seems to only apply when it comes to popular initiatives.
Regardless of what your opinion is on the notion of our modern idealization of liberal democracy, the reality is that for the most part, it does not really exist aside from in our own imagination. To the extent that the average person gets anything they want, this is mostly just a coincidence driven by the fact that oftentimes, the average citizen and the affluent citizen agree on a policy issue. Even though Switzerland is a standout case because popular initiatives seem to give the average citizens a voice, at the end of the day, this only matters if you think equal representation is a good thing. Elitism isn’t inherently bad, and in most democracies, voting, the proud tradition that defines democratic societies, is already somewhat selected. In the United States, non-voters have lower income, lower education levels, are younger, have less trust in the system, and are less likely to identify with either of the two main parties (Pew Research Center, 2014; Thomson-DeVeaux et al., 2020). There is evidence to suggest that people who are more politically towards the center are less likely to be politically literate compared to radicals (Pew Research Center, 2021). A similar case can be made for Swiss democracy. It is exceptional for one simple reason: low voter turnout, which holds true even after accounting for its large population of foreign nationals:
U.S. voting-age population turnout is still behind many other countries despite its recent rise, though registered-voter turnout is remarkably higher
Pew Research Center (2022)
Who votes and who doesn’t on the national level in Switzerland? Well, Blais (2014) found that compared to Germans, the Swiss are less likely to vote if they perceive government and politics as complicated, indicating clear evidence of selection.
We also have evidence that granting women suffrage resulted in increased welfare spending throughout Western countries (Kirkegaard, 2019). This should not be surprising at all, as democracy represents the lowest common denominator of all its constituents, so when the composition of its constituents change, we can predict what the effect will be based on the changes in constituency. The truth is simply that democracy is not some wonderful magical system. We should not expect perfectly equal representation between various groups in society, nor is that a desirable outcome we should be aiming to achieve. Elitism is the norm and will likely remain that way for the foreseeable future. The real problem is not that elitism exists, it’s the fact that the elites are delusional and out of touch with reality. People are only starting to feel discontent with the state of “democracy” because the disconnect between elite and average interests are growing increasingly apparent. In Charles Murray’s book Coming Apart: The State of White America, he presents evidence of a divergence which has occurred since the 1960s among classes. Affluent Americans have grown ever less likely to have interacted with Americans of lower classes, instead living in their own gated communities and ignorant of the real world around them. All the while low income Americans see their situation deteriorating as time passes by, the elites become ever more out of touch with the rest of America. It would be interesting to see where the future leads to once the elites and the average citizens have grown so far apart in their preferences, that even coincidental “victories” on policy issues for the latter become a rare occurrence.
