Skip to content

Who spends how much, and on what?

Nathan Yau (link from Dan Hirschman) constructed the above excellent visualization of data from the Consumer Expenditure Survey. Lots of interesting things here. The one thing that surprises me is that people (or maybe it’s households) making more than $200,000 only spent an average of $160,000. I guess the difference is taxes, savings (but not retirement savings, as that’s included in the “Personal Insurance and Pensions” category), and charitable donations. (Here’s the definition of all the expenditure categories.) Still, I was surprised that the amount spent by upper-income people was so low. The average income of people who make more than $200,000 must be pretty high . . . Here’s something from Wikipedia: of the households making over $200,000 in 2014, average income is (2.61 * $220,000 + 3.28 * $400,000)/(2.61 + 3.28) = $320,000. So, according to the data, I guess that half their income is going to taxes, non-retirement savings, and charitable donations. Put that way, this sounds about right: depending on where you live and how much you make, taxes can take close to 50% already.

Yau reports that he “made these charts in R with a variation of this unit chart tutorial.” He gives links to these, but I’d really like for him to share his code! I’d also like to know exactly what table he used. Yau’s blogging for free so I’m not complaining at all; I think his graph and follow-up discussion are great as they are! and they’d be even better with links to data and code.


  1. Briton Wells says:

    I always find these types of analyses interesting. The lifetime value of extra income varies greatly depending on the income you already earn. Going from 30k to 40k in income may just mean spending a bit more on housing and food, while going from 100k to 110k means that (very likely) the entire 10k will be invested and grow for decades

  2. I find these stacked things hard to scan compared to, say, 10 overlaid lines (or even 10 separate plots). Mainly it’s hard to read the percentages off for any cell. And it’s hard to read the trends with the y axis baseline changing for each category. And they don’t give any indication of what I suspect to be huge variation in these categories among individuals.

    Is the y-axis after tax spending? Is it adjusted by averaging the amounts or the percentages across individuals? It sounds like the former, but the axis labels are unclear.

    @Briton Wells: Going from 15K to 25K also means less debt every year. The graph shows that jumping from 100-150 to 150-200 has very little effect on percentage of income put into retirement (though it’s hard to tell with these charts!). So it looks like 20% of the additional income rather than 100% gets dumped into savings.

    • Another issue is what constitutes “spending”. So if you contribute to a 401k I suspect that goes into “Personal Insurance and Pensions” and this quantity is capped at something like 15% which is suspiciously close to your approximated 20% number. But if you take a bunch of cash and dump it into a regular brokerage account, I suspect this is not “spending” and so it becomes part of the mystery of Andrew’s question about where is the rest of the income going? At the same time, growth of the savings is not going to show up as income until you sell the stock and spend it… which winds up adding a lot of noise because *individuals* particularly in the high income groups move from one group to another depending on how they manage their investments. In fact there are plenty of very wealthy people who have zero wage income (and pay zero FICA tax for example).

      Also, savings is often for the purpose of some future spending… like on kids education or buying a car or boat or house or starting a small business or whatever. That gets into the heterogeneity on the individual level you’re talking about.

      • Thanatos Savehn says:

        A point comes when you can’t rationalize yet another jetski for the dock at a house you visit maybe once a year. Thereupon it becomes a game of Monopoly; unless you’re lucky enough to find ideas with which to (intelligently, hopefully) engage – rather than seeing how many orange notes you can accumulate in whatever time you have left in this weird game we play.

        Now, that said, back in the day when I was young I represented a housing authority and had the chance to witness up close the pathologies of the poor. The saying, “Poor folks got poor ways” is a model well-tested and well-fit. They often live a Hobbesian nightmare and giving them $40,000 or $400 per year will in most instances do little more than enrich those who prey upon them. There’s something (good) to be said for paternalism in such cases.

        • It’s often extremely expensive to be poor. I mean this in terms of unit costs. If you have only a few dollars you must spend it on very small quantities of things at nearby stores because you can’t afford transportation. This leads to doing things that seem illogical to those with more income. Poor people have skills adapted to inefficiency like this, and often won’t have skills adapted to improving efficiency once more money is available. The solution is education rather than paternalism, teach a man to fish and all that…

          • There is also the issue of mental illness, a reason why some people are poor, in that situation a certain level of paternalism is probably a good thing, assigned case workers and money paid straight to rent rather than bank account and soforth. My sister does mental health work and the need for paternalism when people are not able to percieve reality correctly is very obvious.

    • jrkrideau says:

      I find these stacked things hard to scan compared to, say, 10 overlaid lines (or even 10 separate plots)

      This one is a lot better than most stacked charts but I am in agreement with you. I was thinking dotcharts perhaps.

  3. This is almost certainly households rather than people (and some households have just one member of course).

    I’d really like to see this for households with children. There’s something very different going on there I think, food and beverage, education, transportation, healthcare… those are all going to be very different for households with children I suspect.

    And Briton Wells’ point above is also good. The major component that changes across the spectrum here is insurance and retirement savings. This will result in upper income families breaking away from lower income families through exponential growth in their investments.

  4. Lord says:

    I always find spending and saving rather ambiguous, saving in good part being pre-planned spending which when it does occur can easily lead to spending in excess of income for that period. It can also be ambiguous whether spending or saving like housing, a current expense, a stream of services, and a saving/investment.

    • The importance of these issues also varies a lot across income spectrum. Obviously people in the bottom groups probably don’t have much savings, and yet are able to consume much more than they make if they qualify for subsidies (like food and housing, healthcare, and childcare subsidies).

      Perversely, people can often only afford children if they make very little or quite a bit. For example suppose a single woman has a lower middle income (like say $40k) and has say unexpected twins, then she’s likely to be driven towards zero income, and live off housing subsidies and SNAP and soforth. The cost of child-care is too high to make it profitable to work for $40k and eligibility for subsidies increases with decreasing income. I’ve seen graphs that show that maximizing consumption for people like this whose feasible income is less than $45k occurs around $10k income in Illinois, and total consumption drops off a cliff around 20k income… there’s a big “potential energy gap” between 20k and 45k where you’re working harder and doing worse. There are lots of complicated issues involved surrounding policy specifics though.

      • Here’s the PDF with tons of graphs of total consumption vs earned income for various welfare and work scenarios in various counties. I’m sure you can find fault with the specifics of some of this, but the fundamental qualitative result that a single mother should probably not try to make more than 10-20k a year unless she’s able to make more than $70k is likely correct.

        • Carlos Ungil says:

          You said the other day that “any family with less than 90th percentile income would likely wind up better off” with your universal income proposal. I assume this single mother making $20k and getting $40k in benefits would be among the unlucky ones?

          • It’s hard to say in the long term because many things are going to change under that kind of policy situation. But ultimately, I suspect no she’d wind up better off, because many more options open up to her and the overall economy growth improves a lot. Thinking just about her without the broader effect on the country-wide economy would miss the point, though in the short term there may be bigger issues.

            Remember in my policy proposal we have a variety of free parameters we can tune, though not as many as the current system with it tons of special rules.

            UBI for the adult. UBI for the children. Non-income based modifications to the UBI based on: health status, participation in education, age, anything that adjusts the UBI not based on income is “fair game” policy wise in that it doesn’t create the low or even negative slope of consumption vs income. Also there’s a stronger incentive to form a cooperative household rather than remain a single mother. For example, if she marries a working man under current scenarios they lose all the eligibility.

            An *all else equal* comparison is inappropriate because the very purpose of UBI+flat tax is to make all else change for the better too on average.

            But even suppose that tomorrow I wave my wand and put her back to her old job with her two kids and flat tax plus UBI… Suppose we use UBI amounts of say $500/mo for her and $250/mo for each child, and 30% flat tax, then we gut the individual programs she’s getting under this example. Remember she had a $40k/yr earnings potential prior to the unexpected children… Post UBI+flat tax proposal we can calculate:

            $12000 in UBI + 0.7 * 40000 in earned income = $40,000 after tax. That’s the same as what she’s be consuming under the example RTC+Cash+Food+Housing subsidy case in the graphs I linked.

            The long term effect of large changes in policy are hard to figure out without careful analysis. I did actually make some effort to put together that careful analysis, but it amounted to a full time economic research job and unfortunately I’m not a full time paid economic researcher… I’d love to have better more complete answers for you, so if you happen to know a policy center willing to hire me full time to work on it using detailed microdata from the ACS and the Consumer Expenditures Survey let me know ;-)

            • Also, it probably shouldn’t surprise us that someone who is maximizing the currently available benefits might decrease their overall consumption post change in policy. But there are also other people to consider. For example the homeless people living in tents on every pedestrian overpass or bridge abutment here in LA County. We have something like 60,000 homeless people in this county, it’s tripled in the last 10 years or something like that, and one suspects they’re not getting much in the way of benefits. A UBI + flat tax could dramatically improve their lives… ultimately any policy evaluation should really focus not on one small population but the overall tradeoffs among all the populations. One major benefit is the potential to lower the barrier to hiring people, as it’s now trivial to administer: collect 30% of their pay and send to the govt, instead of “hire an HR person or service to handle your complicated payroll requirements”

              I strongly suspect that the pie gets bigger with the UBI+Flat tax and that it spreads more evenly among the population compared to the “rich get richer and poor people can’t ever save or invest or risk losing benefits” scenario current policy creates.

              I have a friend whose mother in law is physically disabled. He’s told me that if she has more than something like $2k in her bank accounts she loses her disability income. That sort of stuff must be affecting millions of people very adversely. $2k doesn’t even fix your air conditioner in your car.

            • Anonymous says:

              > $12000 in UBI + 0.7 * 40000 in earned income = $40,000 after tax. That’s the same as what she’s be consuming under the example RTC+Cash+Food+Housing subsidy case in the graphs I linked.

              Working twice as much to get the same consumption doesn’t look as an improvement. And for some reason you left out half of the benefits. Your comparison is missing $20k in subsidies (Child Care + Medical Assistance). Until the long-term comes and everything gets so much better via some undefined mechanism it seems she winds up worse off.

              • As I say it’s hard to say what happens short term, but the all-else equal fallacy is severe if you try to do this calculation as I did. The point of my calculation is just that it’s not necessarily the case that she winds up dramatically worse off, like she would if for example you eliminated all the subsidy and let her just take home $17,000 of the $20,000 she’s earning because she took a lower paying job to get the subsidies.

                As for “undefined mechanism” the mechanism is simple: avoiding the scenario where people under-utilize their earnings potential to maintain eligibility for benefits. In the scenario where a person who could be earning $40k chooses instead to earn $20k in order to maximize consumption after benefits, the country as a whole loses $20k in potential earnings per person doing this… If it’s even say 10 million people doing something like that we’re talking about loss of productivity equal to something like 20% of the discretionary spending of the entire US Govt. But lost productivity also applies to people who aren’t benefits eligible, because households with higher incomes experience very large marginal tax rates also discouraging use of earnings potential.

                Elsewhere on this thread I’ve shown how there are mechanisms that can replace this hypothetical women’s benefits without creating problems in terms of disincentives to work. For example making the UBI for children age dependent and large enough for pre-school children to afford pre-school attendance (and conditional on their attendance)… across the board regardless of income. That has huge implications for child welfare, and allows us to recoup these huge lost productivity issues.

              • For example if you make the first 5 years UBI for children equal to $2000/mo if she puts them in pre-school the mother’s take home income becomes

                $6000 (herself) + $24000 (children) + $40000 * .7 = $58000 which is basically the same as if she were earning $20k and consuming every last benefit they could think of.

                And, by the way, value of goods and services provided increases by at least $20k because she’s doing a $40k job instead of a $20k, plus the net effect of increased employment in the child-care sector, minus the effect of decreased government administrative jobs. We need to look globally to see how the whole thing plays out, but no there’s no reason to think that necessarily single mothers with two kids and middle income potential necessarily have to be harmed under feasible policies.

              • Carlos Ungil says:

                > + $24000 (children)

                You would be giving the same amount to every child of eligible age in the US, regardless of household income, right? Maybe the fixed 30% tax margin is not enough to pay for that.

              • Whoops, sorry, 2k/mo/child in preschool and 2 children so annual income is:

                6000+48000+0.7*40000 = $82k/yr which is more than the report for the hypothetical current income maximizer.

                Of course we need to analyze globally to figure out whether 30% flat tax is sufficient here, but my point is that there’s not necessarily a reason to believe that this woman *must* be hurt by the proposal of flat tax + UBI + adjustments to UBI based on circumstances other than income.

              • Carlos, yes you’re right, but also in this case we’re globally making $20k more goods and services by employing this woman more efficiently. And there are potentially lots of *other* people we can employ more efficiently as well, such as a spouse of a $120k/yr earner who has 2 children and $90k earnings potential, but after almost 50% marginal tax according to the current taxation system plus after tax cost of a nanny or child-care is staying home doing what is maybe a $30k job of caring for the two kids instead of creating that $90k in value.

                Or whatever. The analysis and feasibility is impossible to determine without a good quality global economic analysis based on actual data from the ACS… something that is very doable, but not on most people’s radar.

              • Here’s a recent article that suggests that this isn’t just a minor problem affecting a few low income women:


                High-income-potential women are finding that they have a hard time utilizing their relatively increased earnings potential, and wind up staying home with kids. This article doesn’t put the finger on high marginal tax rates as the cause, but I think that’s just because it’s not quite on the radar. It’s not the *only* cause of course, but it is a major contributing factor, so again we have lost productivity which is the real cost of current policies.

              • Carlos Ungil says:

                > is staying home doing what is maybe a $30k job of caring for the two kids

                If staying home caring for two kids is “maybe a $30k job”, why would you give the mother in your other example $42k (tax-exempt) extra if she puts her two children in pre-school?

              • Carlos Ungil says:

                (Ok. I’ll stop now. I promise.)

              • Because pre-school has benefits for children in terms of educational outcomes? Also parents should get something to help their kids even if they don’t send their kids to pre-school, and also these are all just example numbers which are only good to 1 significant figure?

                I’ve seen it myself in terms of the effect of pre-school on elementary school outcomes. Kids with pre-school experience can be several grade levels ahead in terms of skills compared to kids without. Kindergarten and first grade often involve basically just keeping some of these kids from hurting other kids instead of learning to read or whatever. That gap builds through time so that by say 9th grade kids without pre-school experience may be performing at levels that kids with pre-school experience were performing at in 6th grade…

                But you generally don’t get that from stay-at-home nanny or mommy care. It’s hard enough in a 1 on 1 situation to just keep your kid from screaming. Pre schools are generally great in my opinion.

                All I ask is that having read all this discussion, you hopefully understand that there’s *something to this idea* and that it *does* require a global type analysis not just a few example gotcha numbers, and that the change is large enough that *all else equal* is the wrong way to think about it. If you can acknowledge that, then I think we’re in the same boat. I don’t have the hard data to prove that the idea works dramatically well, but I have approximate quantitative back of the envelope calculations based on aggregate numbers which suggest that we should put more prior weight on benefits.

              • Also note that there’s a tendency to calculate this stuff as if it’s going to be a balanced budget… income from taxes equals expenditures and transfers by the government. But the current system is *not* a balanced budget, and so that’s the wrong way to think about the problem. The more apples-to-apples way is to keep the growth rate of the debt unchanged.


                allows you to get some aggregate data on the population by age.

                US GDP in 2016 was 18.6 Trillion, or about 58k/capita (man woman *and* child)

                So a 30% tax on all income = 17.4k/person. There’s about another 1800 in deficit spending… let’s round off to $20k/person to spend. Suppose operating essential functions of govt costs around the same as the current discretionary spending: $1 Trillion, or $3k/person. So we’re left with 14k/person to handle what’s currently handled by a combination of social security medicare and maybe housing and education dept.

                Now on average for people between say ages 20 and 50 you’re spending maybe 6k/person for UBI, and that’s around 150M people, this leaves 3.6 Trillion to spend on the other ~150M people which is an average of $24k/yr/person. So people between age 0 and 20 and age 50-100 can get $24k/yr in UBI… The aggregate numbers are not completely out of whack. We can then refine it some from there… and that’s all assuming *no increase* in GDP productivity. If instead you imagine say an average of 10k/yr/person increase in productivity… that’s an additional 3k/yr/person in tax you can then allocate to the whole scheme… Back of the envelope it’s not unreasonable to consider actually doing the detailed analysis, you can’t rule out it working using this kind of back of the envelope stuff..

              • Never do arithmetic in a blog post… While editing things I went back and checked certain numbers and then altered them, without altering the numbers that they later implied… 20k-3k = 17k not 14k for example. so it becomes even more potentially workable… This kind of stuff can be done better in a spreadsheet anyway, it actually is an appropriate use of spreadsheet technology. But if you do it, you don’t wind up dramatically out of whack, the back of the envelope calc doesn’t immediately spoil the fun.

            • Rahul says:

              >>>Remember in my policy proposal we have a variety of free parameters we can tune<<<

              So how does one tune them? What's the objective function here?

              • That’s a good question. A very good question. Obviously it’s politically charged. But one could imagine that maximizing GDP and minimizing say suicide, malnutrition, violent crime, and family bankruptcy rates could be somewhat low-controversy.


              • Under that kind of utility function for example it might make good sense to set the UBI amount per child under 5 years old to high enough that they could all participate in approved types of preschool, and make it conditional on participation. Here’s a good way to invest in child health and welfare, and yet it doesn’t create an income disincentive. It also gives stable childcare enabling mothers to work, without the need for the government to *run* it.

              • Rahul says:

                Isn’t this again a kind of “mathiness”? The way you’ve set up the model, including the knobs, tuning and all it seems all nice & mathematical. Which is great.

                But at the end of it we anyways will decide the thresholds of these parameters relatively arbitrarily in response to ad hoc political pressures?

                Does anyone ever set GDP as a objective function and optimize the model and let the parameters fall where they may?

              • I don’t think we need to optimize an objective function using a fancy algorithm here. What we need to do is create a system which is simple enough that it doesn’t accidentally create perverse incentives. None of the individual programs designed to help a single mother with 2 kids are saying “Hey you know what, if we make it so that 20k is the optimum income we can lock women into poverty forever and ensure that they are fundamentally reliant on our program thereby also ensuring long term job stability for our administrators” but that’s kinda what they do.

                Of course the whole thing evolves according to a politically motivated random walk, but if the rules prevent you from creating downward sloping Consumption vs. Income curves at least you won’t create poverty traps and dramatically destroy productivity all hidden under the blanket of “helping poor mothers” or whatever.

              • One of the problems is pretending that dollars changing hands is the same as costs. In economics all real costs are opportunity costs. If we tax a dollar from Rahul and give it to Me, I will spend the dollar and Rahul won’t. From a global perspective there’s not any real “cost” here. Of course in order to run this program we’ll have to employ some tax collector and they’ll need to take some percentage, but they are going to spend their percentage as well… so again no real cost.

                But as soon as Rahul decides based on this policy “f*** that I’m not going to bother making that extra dollar”. NOW we’ve got a *real* cost because guys who want to pay Rahul to design skateboard wheels or whatever don’t get their skateboard wheels designed because he decided to stay home and watch TV…

                So the goal in my suggested policy was to dramatically reduce the real cost of the policy by always ensuring that there is an incentive for people to maximize their productivity because they always take home some significant portion of every marginal dollar they could earn, while still ensuring that people are receiving sufficient income to survive and be healthy and not create other kinds of societal costs (health care, mental health, violent crime, opioid abuse, living on the streets etc).

            • The part about incentives to form a cooperative household is really nontrivial. Housing has been an increasingly large cost for families. I suspect the graph here averages over many years and so may not really show that.




              Paint a different picture than this graph here for example.

              Suppose the single mother with two kids moves into a shared 4 bedroom apartment with another symmetric single mother with two kids, and splits the rent? Household income after taxes goes to something like $80k, and rent and food expenses can be shared efficiently. But try doing that in “project” housing.

              etc. The all-else-equal fallacy is easy to fall into ignoring that under the two policies the available options and the outside economic conditions potentially change dramatically.

              • Rahul says:

                >>>Perversely, people can often only afford children if they make very little or quite a bit. For example suppose a single woman has a lower middle income (like say $40k) and has say unexpected twins, then she’s likely to be driven towards zero income, and live off housing subsidies and SNAP and soforth. <<<

                Aren't many of these perverse effects simply because we use thresholds instead of continuous functions?

                Imagine we kept the major policy the same but replaced every step-change threshold with a continuous function wouldn't that get rid of many of these perversities which are essentially a gaming strategy to drop below an arbitrary threshold to make benefits magically kick in?

                If instead a delta decrease in income got you delta more in subsidies would people be as motivated to use such market distorting behaviours?

              • Thresholds are stupid I agree, but the bigger issue is that consumption vs income must always have positive slope, and probably something like positive 0.5 or more at all locations. This slope is 1- the effective marginal tax rate. There are people who have effective tax rates of well over 100% due to policy issues. Even at 50% you destroy a lot of productivity potential.

  5. Matt says:

    According to the definitions you linked, charitable donations are already included in “Cash Contributions.”

  6. Nat says:

    I don’t like this visualization. I think it is confusing and doesn’t really tell a story. The main story seems to be growth of the “Personal Insurance & Pensions” group. If that group was broken out as a bar chart then this would be very obvious. I also suspect that there might be a better way to arrange the groups to be more informative. For example, placing the “Personal Insurance & Pensions” group at the top of the chart might help. It is clear that group is growing, but it is not so clear which group(s) are shrinking besides “Housing”. I think a different chart for each group (e.g., facet_wrap with free y-scales in ggplot2) might be clearer.

    I am interested in the correlation between income and proportion of spending in each category. Perhaps several scatter plots would be better?

  7. Dzhaughn says:

    The education/entertainment ratio is a bit embarassing across the spectrum. The education and anything else ratio, really. Is “edutainment” the new thing anymore?

    When I drive 20 miles out of town to a farmer’s market to buy 3 lbs of excellent green beans and tomatoes at $1/lb, how shall we count that spending? I guess $6 of transportation, $3 food. Say $1 for the Dan Simpson spotify playlist I’m “singing” along to, file under “education.” Or I can have it delivered by Whole Foods for $8, file under food (because the cost of moving food is food, not transportation) so I can watch 100 episodes of Penny Dreadful (ack phthfptt) for $1 on netflix.

  8. peter says:

    I’m much more interested/depressed by the pattern visible in the lower income brackets than the upper. It looks like the average person making under 40k/year is accruing debt every year.

  9. Doug says:

    FYI this visualization seems similar in style if not in substance to what Kaiser Fung calls “mosaic” charts or others call “marimekko”. See and And an example with ggplot2 is here:

Leave a Reply

Where can you find the best CBD products? CBD gummies made with vegan ingredients and CBD oils that are lab tested and 100% organic? Click here.