The Daily Rhythm

It’s no secret that I’m a bit obsessive about my news habits and filters, manifested most obviously in constant tweaking of my RSS feeds but also in who I follow on Twitter, what email newsletters I allow into my inbox, etc. The challenge in these adjustments is always to strike a balance between drinking from a firehose and a strange kind of fear of missing out. Call it the Goldilocks news filter problem: how much is just right?

Today, the challenge is much more about avoiding the allure of too much and too often rather than too late or too little. Like BuzzFeed’s Charlie Warzel, I recently went on a big unfollow craze with Twitter, and even beyond that use private lists to further filter out the noise. But more so I’m finding lately a real increased affinity for email, as well as an undiminished love for RSS. Specifically, I’m finding that daily is, in almost every case, a regular enough cadence for following the topics I care about.

Part of this is professional. Last summer I moved from a job that included following breaking news and writing about it as quickly as possible to one that is less dependent on the latest developments. But that’s just a piece of it. Even for topics that I don’t write about, I’m finding daily to be roughly the right approach.

Every morning I read three email newsletters and sort through 15 RSS feeds, mostly from media organizations, and another couple dozen feeds that are more infrequent from individual bloggers. Some of it I simply skim the headline and mark as read; some I open and read; some I save to Pocket for later.

Sure, I browse Twitter at points throughout the day, but less than I used to. Mostly, I’m finding, things can wait until the next day.

How will humans add value to algorithms?

Last month, I wrote a piece at HBR about how humans and algorithms will collaborate, based on the writings of Tyler Cowen and Andrew McAfee. The central tension was whether that collaboration would be set up such that algorithms were providing an input for humans to make decisions, or whether human reasoning would be the input for algorithms to make them.

One thing I thought about in the process of writing that piece but didn’t include was the question of whether one of these two models offered humans a higher value role. In other words, are you more likely to be highly compensated when you’re the decider adding value on top of an algorithm, or when you’re providing input to one?

I was initially leaning toward the former, but I wasn’t sure and so didn’t raise the question in the post. But the more I think about it, the more it seems to me that there will be opportunities for highly paid and poorly paid (or even unpaid) contributions in both cases.

Here’s a quick outline of what I’m thinking:

Screenshot 2014-01-26 at 3.26.26 PMIt seems totally possible for the post-algorithm “decider” to be an extremely low level, poorly paid contribution. I’m imagining someone whose job is basically just to review algorithmic decisions and make sure nothing is totally out of whack. Think of someone on an assembly line responsible for quality control who pulls the cord if something looks amiss. Just because this position in the algorithmic example is closer to the final decision point doesn’t mean it will be high value or well paid.

Likewise, it’s totally possible to imagine pre-algorithm positions that are high value. Given that the aggregation of expert opinion can often produce a better prediction than any expert on his or her own, you can easily imagine these expert-as-algorithmic-input positions as being relatively high value.

Still, the onus is on the experts to truly prove useful in this scenario. Because if they’re not adding discernible value, waiting in the wings is the possibility for the algorithm to aggregate an even greater range of human judgment — say via social media — that could be done cheaply or even for free.

I’m not sure where this leaves us except to say that I don’t see much reason for us to be “rooting” for algorithms to be inputs to humans or vice versa. In all likelihood this is not the right question. The relevant question, and a harder one, is simply how do we apply human judgment in a way that enhances our increasingly impressive computational decision-making powers.


iPhones and Income: Does Technology Change the Middle Class Stagnation Story?

Steve Jobs Announces the iPhone in 2007

One of the most common responses to my post on middle class incomes was to point out the role of technological progress. If the average American family went back in time to 1989, I wrote, they’d make just as much money but work longer fewer hours to do it. But, some responded, they wouldn’t have iPhones. That isn’t meant to sound trivial, and as someone optimistic about technology I don’t consider it to be. Improvements in technology are an important piece of any conversation about progress. But do they change the story about middle class incomes?

Yes and no.

Short version: All of the data I included adjusted for inflation, which accounts for certain kinds of technological progress but not others. Some new technologies – like the iPhone – aren’t currently captured in that data. Others are. If new technological inventions like the iPhone were able to be included in common inflation measures, the incomes of the middle class would indeed look at least a bit higher.

Here’s the long version, starting with a short overview of inflation.

Measures of inflation track the price of goods over time, and although it’s technically an oversimplification, you can think of such measures – like the Consumer Price Index (CPI) – as a proxy for the cost of living. If the stuff you need to get by costs, in total, $100 per week today, but next year that same stuff costs $200 per week, you’d need to be making twice as much money just to be keeping up. So if you hadn’t gotten any raise over the course of that year, an inflation-adjusted (“real”) accounting of your income would say that your income dropped 50%. Inflation-adjusted income measures account for how much stuff costs.

Prices don’t all change together of course, so the CPI uses a bunch of “baskets” of goods. Food is one part of that. Let’s say the price apples goes up, but the price of bananas goes down. If those changes average out, from the CPI’s perspective, “prices” haven’t changed. (If this happened, you might choose to buy only bananas for a while, in order to take advantage of the low prices. So this is an example of when the CPI starts to diverge from cost-of-living. That’s called the substitution effect and it’s one of the big challenges to measuring inflation, but it’s a bit outside the scope of this post.)

In theory, technological improvements should be captured in measures inflation. Say one of the things most people do is to send letters, documents, and other information to each other. It used to require going to Staples, buying envelopes, paying to print, then paying for postage, etc. Now you can just email them from a relatively inexpensive computer in your home. The price of sending all this stuff, one of your regular life activities, just got cheaper. Inflation is about measuring prices, so a measure of inflation should capture this price decrease. If the inflation measure captures it, it would mean that inflation-adjusted income (like I used in my previous post) would capture the impact of tech.

But in practice, measures of inflation have a really hard time capturing new technologies. To see when inflation does and doesn’t capture technology, let’s go back to the food example.

The kind of technological change that inflation is relatively well set up to track is the kind that results in decreased prices for an existing good. Say a farmer comes up with a new way to grow apples and the result is that the exact same kind of apple you’re used to buying suddenly costs half as much as it used to. The CPI will capture that decrease, and so inflation-adjusted income will reflect the improvement.

But say an agricultural scientist invents some new health shake, unlike any food out there on the market, which provides all your daily calories and nutrients. This counts as a “new good” and inflation measures don’t really have any way to account for it. In practice, if a bunch of people start buying the health shake, after a while the Bureau of Labor Statistics will decide to add it to the CPI and start tracking changes to its price going forward, but this misses the value of the new invention in two respects.

The first, and simpler, problem is that the BLS only updates the CPI’s “baskets” every four years. And for some technologies, prices can drop a lot over that amount of time. So imagine the health shake debuts at $100 per serving, but four years later, by the time the BLS gets around to counting it, it’s going for $20 per serving. That price decrease will be missed.

The second issue is a trickier. The very act of invention, if the new product is novel enough, is simply not accounted for at all in inflation statistics. Here’s how a report from The National Academies puts it:

Without an explicit decision to change the list of goods to be priced, standard indexing procedures will not pick up any of the effect of such newly introduced items on consumers’ living standards or costs…

…If significant numbers of new goods are continually invented and successfully marketed, an upward bias will be imparted to the overall price index, relative to an unqualified [Cost of Living Index]…

…Proponents of more traditional price index methodologies argue that it is a perversion of the language to argue that the effect of, say, the introduction of cell phones or the birth control pill is to reduce the price level, a result that comes from confusing the concept of a price level with that of the cost of living. Their position is tempered somewhat by the realization that, outside of price measurement, there is nowhere else in the national accounts for such product quality improvements to be included and, as Nordhaus (1998) and others have argued, real growth in the economy is thereby understated.

How would the introduction of a brand new good be translated into a change in price? The idea here is that sometimes a new good comes to market at a price lower than some consumers would have been willing to pay. Our magic shake example comes to market at $100 per serving; but perhaps some consumers would have been willing to pay $200 per serving for it, but just never got the chance because the technologies that make it possible hadn’t yet been invented. This difference represents value that inflation measures won’t catch. (An interesting note for innovation econ nerds: this is less likely to be a problem to the extent you see technological innovation as a demand or “pull” driven process. It’s really supply shocks that will cause big problems for inflation measures.) There are econometric techniques that some experts believe could be used to capture this value, but they are complex, controversial, and not yet in use.

To sum up, here’s how to think about it: when Amazon uses better software to make retail more efficient and therefore makes a bunch of consumer products cheaper, that’s captured in our most common measure of inflation. But when a radically new consumer product — like the iPhone — is introduced, some portion of the new value will go uncounted. If the iPhone gets cheaper over the first few years before it is incorporated into the CPI, that value will be lost. But once it is included, improvements in technology that make the iPhone cheaper will be captured.

The result is that inflation-adjusted income measures do fail to account for certain kinds of technological progress. How big is that bias? Best I can tell, we don’t really know. Some have suggested it is sizable, but there is no consensus.

So as for the response — sure, middle class incomes were the same a decade or two ago, for fewer hours worked, but now we have iPhones — it is on to something. It’s perfectly reasonable to point out that certain new tech products are available now and weren’t then, and that income data doesn’t fully capture that. But be careful with this argument. It’s not all new tech that goes un-captured. Lots of the behind-the-scenes increases in efficiency due to tech that result in lower consumer prices are captured, as is at least a portion of the continuing decreases in price for consumer tech products once they’ve been in the market for a while.

So it’s a good point, but a nuanced one.

UPDATE 2/5/14: Martin Wolf at FT nicely captures this in two sentences: “Its price was infinite. The fall from an infinite to a definite price is not reflected in the price indices.”

Middle Class Incomes: Up, Flat, or Down?

Screenshot 2014-01-06 at 4.27.09 PMDepending on who you ask, the incomes of the American middle class over the past few decades have either a) risen only a little b) stagnated, i.e. stayed flat or c) declined. When President Obama declared in his State of the Union speech that family incomes had “barely budged” from 1979 to 2007, The Washington Post called it inaccurate, noting that median household income increased substantially over that period. And yet barely a day goes by without a story that references stagnating wages for the middle class.

So which is it?

The one thing everyone agrees on is the fact that the rich are getting richer much, much faster than anyone else. And so in one sense, it doesn’t much matter if the answer is (a), (b), or (c). In either case, rising inequality represents a gross misallocation of the nation’s wealth. Still, the possibility that the average American is worse off economically than one or two generations ago makes the issue feel all the more urgent.

Unfortunately, the seemingly simple question of whether Americans are making more money today than in decades past is a bit tricky. The answer depends on the timeframe and the measure.

Short version: The average American family was making modest gains in income over the past few decades, but was working longer hours to do it. Then the recession happened and set the average family back 10 to 20 years.

Now here’s the full story.

The easiest place to start is household “market income”, which just means how much money a household makes before taxes or government transfers are counted. (Importantly, employer-based health care benefits are included in this measure.) Here, via the Congressional Budget Office, is the snapshot over time:

market income CBO

The thing to note here is that the median household income rose nearly 20 percent between 1979 and 2007. That the mean income rose faster hints at the fact that the rich got richer at a much faster rate (see the chart at the top), but nonetheless, seen at this level the story looks like one of modest progress.

Things actually look better still when you consider the impact of taxes and transfers, shown here again via the CBO:

median after tax income CBO

Once taxes and government transfers are accounted for, the median American’s income has risen more than 30 percent from 1979 to 2007, making the story of minor progress a bit more progress-y. (It’s this after-tax measure, accounting for taxes and government transfers, that you saw in the chart of all income inequality at the top of this post.)

There’s one more upside to note: since the average household is smaller than a few decades ago, these gains are slightly larger when size of family is accounted for. Unfortunately, that’s the beginning, not the end, of the story.

It turns out that the median household’s income has only increased because that household has been working more. The New York Times summarizes data from Brookings from 1975 to 2009:

Median wages for two-parent families have grown 23 percent since 1975, after adjusting for inflation. The collective number of hours worked by both parents over the course of a year, however, has risen 26 percent. That means their wages haven’t even grown as much as their working hours would imply they should.

The increase in hours worked is largely the impact of women entering the workforce. To make that point a bit more clear, we can look at this chart from Brookings:

Screenshot 2014-01-06 at 8.23.10 PM
If modest increases in household income for the median family are the result of more hours worked, what do wages look like on an hourly basis? For that we can turn to the Economic Policy Institute:

Screenshot 2014-01-06 at 8.17.41 PM

As you can see, the story is one of stagnation since the 70’s, with a modest boost in the late 90’s. This is what stories about “stagnant wages” are talking about. The average American doesn’t make much more for his or her time than in the 1970’s. To bring in more income requires working longer hours.

But here’s where it goes from depressing to downright infuriating. That modest increase in household income that the median family earned by working longer hours? Well, not surprisingly, the Great Recession pretty much wiped it out:

 since the recession
As for the recovery? Well, in its first year, from 2009 to 2010, the top 1% captured 93% of the income gains, according to Stanford:
Screenshot 2014-01-06 at 8.34.41 PM
What about since then? According to Pew, the top 7% saw its wealth (not income) rise 28 percent from 2009 to 2011 while the bottom 93 percent of Americans saw their net worth decrease by 4 percent.
Screenshot 2014-01-06 at 8.37.45 PM

So there you have it. Wages are flat, incomes were up but only because of more hours worked, and then got hammered by the recession. If the average American family could take a time machine back to 1989 they’d make just as much money, and would work fewer hours to make it.

The typical argument as to why we can’t do anything to fix this claims that intervening would jeopardize economic growth. Even if that were true, what’s the good of growth if it doesn’t make anyone richer except the rich? And let’s be clear: that is what economic growth has done.

Here’s where income growth has gone from 1979 to 2007:

Screenshot 2014-01-06 at 8.42.00 PM

But remember: that was the pre-recession distribution. It’s only gotten worse.