As the votes rolled in, one thing became increasingly clear. Hillary Clinton won the presidential election by a nearly 3 million vote margin, but lost the Electoral College by 74 votes—the greatest discrepancy between the popular and electoral votes in American history. The result spawned an uproar. “The election was rigged” and it’s time to get rid of it, the Los Angeles Times exclaimed. Before leaving the Senate, Barbara Boxer of California even introduced legislation that would replace the Electoral College with a popular-vote system.

But despite the opprobrium aimed at it, the Electoral College remains little understood. Here’s a quick rundown: The number of Electoral College votes each state has is equal to the number of senators and representatives it sends to Congress. Each state is guaranteed two senators and one representative, so each state gets at least three electoral votes. Because of the 23rd Amendment, our nation’s capital also gets at least three electoral votes.

Every 10 years, the Census Bureau counts the population in every state and applies a convoluted formula to determine the number of representatives (and therefore the number of electoral votes) each state will receive.

Because small states get three electoral votes regardless of their population, and because the Bureau’s formula prioritizes giving new electoral votes to smaller states, this system creates a predictable bias in favor of small states. And this small-state bias has grown astronomically over time. In 1804, the first presidential election after a census-based reapportionment, the most over-represented state (Ohio) was 90 percent better represented than it would have been if electoral votes were tied perfectly to state populations. In 2015, the most over-represented state (Wyoming) was 306 percent better represented than it would be under a proportionate system.

Not a single state was two times over-represented in 1804. Today, six states and DC get at least two times more electoral votes than they would under an equal-population system, while Texas and California each get 15 percent fewer electoral votes. In 1804, no state was under-represented to that extent. It is hard to imagine that the Framers anticipated the huge disparities in representation the Electoral College currently produces.

But it’s not just that certain states are over-represented. Specific segments of the population get to wield far more power under the Electoral College than they would otherwise. States identified as Republican strongholds, for example, are more over-represented, on average, than states of any other partisan bent.

Gallup has identified 11 “strong Republican” states where support for the GOP outstrips support for the Democratic party by at least 10 percent. These states have 65 percent more representation in the Electoral College than they would if electoral votes were distributed evenly. They also have 12 percent more representation than “strong Democratic” states, at least 25 percent more representation than “weak Democratic” or “weak Republican” states, and 61 percent more representation than “competitive” states. This is largely because the “strong Republican” states of Alaska, North Dakota, South Dakota, and Wyoming all receive two times more electoral votes than they would if votes were apportioned by eligible voter populations.

It’s not just power based on partisan affiliation that gets skewed in the Electoral College system. States that are under-represented in the Electoral College also have larger populations of Hispanic and Asian people. Those groups are, therefore, are under-represented.

Hispanic people are nearly 10 percent less represented in the Electoral College than White people, and Asian people are about 7 percent less represented. The disadvantage experienced by Hispanic people is largely due to the fact that the four states with the largest number of Hispanic people (California, Texas, Florida, and New York) are all at least 8 percent under-represented in the Electoral College. Similarly, the three states with the largest populations of Asian people (California, New York, and Texas) are each 8 percent under-represented in the current system.

Despite the changing demographics in the country as a whole, the under-representation of Hispanic people in the Electoral College is likely to endure. Applying the same methodology to population projections by the Urban Institute suggests that in 2030, 22 percent of the population will be Hispanic, and they will be 7 percent underrepresented.

People in rural, non-metropolitan areas are also given a substantial advantage in the Electoral College based on the edge enjoyed by the states they live in. The average non-metropolitan person is 21 percent better represented in the Electoral College than the average metropolitan person.

Electoral under-representation for members of specific groups, especially specific racial groups, is deeply troubling, not because of what it means for Democrats (who predictably lose out when Hispanic and Asian people are under-represented), but because of what it means for our democracy. First, the founding fathers warned about the tyranny of the majority, and the importance of ensuring that our governmental systems afford effective representation to minorities. The current distribution of Electoral College votes predictably under-represents minorities already struggling for a spot at the table. Secondly, electoral under-representation stems from, and therefore mirrors, under-representation in the House of Representatives. Thus, these minorities are under-represented in both the executive and legislative branches. And finally, as the populations of Hispanic people continues to grow and the bias persists, this two-branch bias will affect a larger and larger number of Americans.

* * *

In 2020, the Census Bureau will set the number of Electoral College votes each state will get in the 2024 election. At that time, if we were to return to the levels of over- and under-representation that existed in 1804, the biases would remain, but they would shrink substantially. Small states, “strong Republican” states, and non-metropolitan people would have less of an unfair edge. Wyoming, for example, would still be 96 percent over-represented (as opposed to 306 percent over-represented under the current system), “Strong Republican” states would be 23 percent over-represented (as opposed to 65 percent), and people who live in non-metropolitan areas would only be 6 percent over-represented (as opposed to 19 percent). Moreover, Hispanic and Asian people would only be 2 percent under-represented.

But there’s a catch. To achieve this Framer’s-era distribution without removing electoral votes from any state, we would have to give many states a lot of additional electoral votes—about 45 to California, 32 to Texas, 23 to Florida, and about 200 combined to other states in inverse proportion to their current disadvantage. Obviously, a sudden addition of 45 Representatives in any one state might be contentious, but we can take incremental steps to approach the Framer’s era distribution of Electoral College representation by making a sizeable increase in the number of representatives and electors in the next election. As radical a solution as it may seem, drastically expanding the size of Congress may be the only way to ameliorate the bias embedded in the Electoral College.

* * *

Democrats keep losing the Electoral College while winning ever-larger popular-vote victories because their support is overly concentrated in under-represented states like California and New York. Clinton won California by over 3 million votes, netting 55 electoral votes. Trump’s combined popular vote margin in Pennsylvania, Florida, Michigan, and Wisconsin was under 250,000, but those victories netted him 75 electoral votes.

Moving forward, Democrats should seek to ensure they not only secure huge victories in very progressive and populous states, but also invest time and resources in securing victories in states where votes are likely to impact the outcome of the election. But which states are those?

Because elections are decided by a vastly different number of voters in different states, and because states have very different numbers of electoral votes, voters in some states are far more likely to impact the outcome in the Electoral College. A voter in a state with a large spread between the two major candidates and/or a small number of electoral votes would have a small likelihood of affecting the election. Think of Washington, DC—an extremely partisan city with a measly 3 electoral votes. Casting your ballot there, for either party, is unlikely to influence the outcome of the presidential election. In contrast, a voter in a state with a small spread between the two candidates and/or a lot of electoral votes can make a difference. Think of Florida, where elections have been decided by a tiny number of voters, but the state provides a whopping 29 electoral votes. Casting a vote in Florida is more likely to swing how the state apportions its huge prize, which in turn is more likely to impact the outcome of the election.

Combining these two factors (major candidate margin and electoral votes), in 2016, voters in New Hampshire and Michigan had the biggest likelihood of influencing the election.

Not surprisingly, the top five states where a vote can make a difference are also typical swing states, and the bottom five are stable Democratic or Republican states. What is surprising is the huge range of how valuable a vote in a given state can be. A vote in New Hampshire is over 120 times more valuable than a vote in DC. While this might seem counterintuitive given that both places have three electoral votes, recall that George W. Bush won by two votes. Moreover, the tiny margin between Trump and Clinton in 2016 meant that a voter in New Hampshire was a lot more likely than a voter almost anywhere else to help their candidate of choice secure any points where it counts.

Moving beyond the top five most influential states, many non-swing-states made the top twenty. A vote in Minnesota mattered 7.5 times more than a vote in the average state, and a vote in Alaska, New Mexico, Delaware, Rhode Island, and Texas (yes, Texas) mattered anywhere from 1.5 to 2.65 times more. Votes in all of these states mattered more than votes in Ohio and Iowa, two typical swing states.

One way to explain Clinton’s loss is that she lost four of the five states where a vote matters the most, but won the two states where your vote matters the least.

* * *

To avoid a repeat in 2020, Democrats should do three things.

First, campaign strategically in states where a vote is more likely to affect the outcome in the Electoral College. While many of these states are swing states, and swing states certainly receive a great deal of attention and ad revenue, there are votes yet to be had in each of the strategic states mentioned above. Even Minnesota, which had the highest turnout rate in 2016, still only saw 75 percent of its eligible voters cast ballots. That leaves a quarter of the electorate. And the average state is far less engaged. Sixty percent of eligible voters nationwide participated in the 2016 election.

Second, overturn or repeal voter identification laws in those key states. Voter identification laws are having a clear impact in key states. They are in effect in New Hampshire, Michigan, Wisconsin, and Florida—four of the top five states with an impact on the Electoral College. And a federal court found that voter identification laws in Wisconsin could be keeping 300,000 voters (or 9 percent of the state’s electorate) from voting, and that these 300,000 voters are more likely to be poor, Latino, and Black. Current data suggests that Trump won Wisconsin by fewer than 23,000 votes. Aggressively taking on these laws is not only a strategic means of providing Democrats with a potential pathway to victory in 2020. It is also a moral imperative for those hoping to remove at least one threat to the franchise experienced by groups that are already partially disenfranchised by the bias in the Electoral College (among other threats to their right to vote).

Finally, minimize the negative impacts of voter registration purges in key states. Both NBC and Rolling Stone have reported that such purges—allegedly designed to ensure individuals are not registered to vote in multiple states or are not registered if they are non-citizens—have a disproportionate, negative effect on the franchise of Black, Hispanic, and Asian voters. This is largely because those populations use fewer surnames and thus are more vulnerable to false-positive identification. Mark Swedlund found that these disproportionate purges are happening in Virginia and Georgia—two of the top 15 states where a vote is likely to impact the outcome in the Electoral College. Ensuring adequate safeguards against hasty purges in these key states will also safeguard the franchise for vulnerable groups while helping Democratic candidates benefit from every legitimate ballot cast in states more likely to impact the outcome in the Electoral College.

* * *

The Founding Fathers envisioned a democracy where all voters could help determine the destiny of the country through free and fair elections. As bias in the Electoral College grows and obstacles to the franchise persist, our nation’s democratic promise fades. To rekindle this promise, and revive fading faith in our democratic institutions, we must restore a semblance of balance to the Electoral College and overcome the laws and practices that are subverting the voices of millions.