Tag Archives: google

Obama Top Chef Romney Founding Fathers


If you don’t collect data on individual web users, and don’t have a big-data budget, you can still learn a lot about how people voted in this presidential election from some creative probing of the Google Correlate database. The power of the tool is in uploading your own data (such as vote tallies) to see what searches mirror your target pattern.

For example, the map on the left is what I uploaded: the ratio of Obama votes to Romney votes in each state, as of Thursday morning. The map on the right, from Google, is the relative frequency of searches for “top chef.” The two patterns have a correlation of .88 on a scale of 0 to 1.

cohen_topchef.pngMaybe it’s a complete coincidence that Michelle Obama appeared on a Top Chef program earlier this year. But out of the 100 Google searches that most closely match that vote pattern, eight are aboutTop Chef. Others on the list include “spliff” (never heard of it), “mos def” and various reggae artists, as well as “itchiness.”

On the other hand, searches for “founding fathers quotes” follow the Romney/Obama ratio just as closely:

cohen_foundingfathers.pngMost of the searches on the top-100 Romney-state list (all correlated about the same .84) are about simple, non-obscene pleasures, such as “clean jokes,” “clean funny jokes,” “funny commercials”; and home-schooling materials, like “flag clipart,” “in god we still trust,” and “printable scrapbook.” After the kids are in bed, though, someone is Googling “hot cheerleader,” before quickly toggling back over to “sean hannity” when he hears mom coming up the stairs.


Filed under Politics

Have Obama haters lost traction?

Maybe it’s because Donald Trump isn’t really a true hero to anti-socialist, anti-Muslim, racist Americans.

For whatever reason, there has been a real slump in the number of people typing “obama gun” (will he take our guns away?), “obama muslim” (the idea used to run at about 20%), “obama socialist” (the republic “hangs in the balance“), and “obama citizen” (thank you, Snopes) into the Google search box since the 2008 election.

Here’s the Google trend (and the search link):

We don’t know how much these fears, versus other concerns, will affect votes against him this year, although there have been some good efforts to track the effects of anti-Black racism on his vote tally.

Naturally, not everyone who Googles these things believes the underlying stories or myths. But it seems likely they either believe them, are considering them, heard someone repeat them, or are arguing with someone who believes them, etc. So I’m guessing – just guessing – that these trends track those beliefs.

But maybe four years of Obama as an actual president has softened up the hard-line hatred in some quarters. What do you think?


Filed under Uncategorized

Word signs of the times

A quick look at some of the trends that help shape (some of) our lives, via Google Ngrams.

As a proportion of references to “theory” and “academic,” uses that begin with “just” are on the rise:

After a lot of talk about “good parenting” in the 1970s, references to “bad parenting” as a share of all references to parenting are gaining rapidly.

As “capital” is increasingly modified, “human capital” took an early lead, but lost it to “social capital,” while “cultural capital” established itself as a third-party niche.

After a tipping point (up) in “tipping point” use in the late 1990s (Gladwell’s New Yorker article was in 1996), will there be a tipping point (down) to end the epidemic?

Finally, to capitalize on all the brouhaha over political forecasting, I offer my own prediction of the rank ordering in this year’s election: Democrat, Republican, Green, Libertarian.

If I am wrong, who cares? I only spent 5 minutes on the forecast — that’s the least of our problems. If I am right, I’ll accept offers for next time.

Leave a comment

Filed under In the news, Politics

Gun Google searches and suicide

A quick addition to the post the other day about gun searches and their association with Mitt Romney, accidental deaths of young people, and divorce rates: suicide.

The Centers for Disease Control has a tool for mapping fatal injuries. It allows mapping down to the county level, with nice options. Here’s their national map for suicide rates:

If you take the list of state suicide rates over to Google Correlate, the answer is, mostly: Guns. I entered the male and female suicide rates separately, and the total rates combined, and guns searches dominate. That is, the searches that are most common where there are more suicides (age adjusted), and least common where there are fewer suicides, are almost all about guns. Again, I’m not expert on the types of guns and paraphernalia, but this is not (just) about hunting: it includes assault weapons, ak-47s, “armor piercing,” “tactical sling,” etc. The correlations between suicide rates and search frequencies across states are high: between .82 and.90. (The full list is below).

I used age-adjusted rates for all ages and race/ethnic groups from the years 2000-2006.

Off the gun subject, there were some interesting other patterns. Both men’s and women’s suicide rates (which are highly correlated, about .90 across states) were strongly associated with searches for the artist Luis Royo, a Spanish artist who specializes in dark, violent and apocalyptic art. Here’s his homepage:

For women’s suicide rates, there was also a strong correlation with searches for “divorce help” and “divorce paperwork.” That’s interesting because suicide is more common among divorced people:

Further, both men’s and women’s suicide rates were correlated with searches for “war footage.” That’s interesting because of the high rate of suicide among soldiers and veterans:

Given how similar the male and female suicide pattnern is, the difference between searches on the male list and those on the female list is interesting, and reveals how sensitive the Google search data are, with millions of searches to sift through. Anyway, maybe someday search patterns can help with identifying risks or contribute to suicide prevention. (Here are my past posts on suicide.)

Here are the complete lists of search terms correlated with suicide rates for men and women across states, in no particular order (all the correlations are similar), condensed a little with the use of asterisks for repeated terms and plurals:


Filed under Me @ work

Guns dividing America (Google edition)

Whenever I get a good indicator broken down by state, I head over to Google Correlate to see how it connects to America’s search behavior. Often what I find is a gun connection. This is very big in searches related to the election, so I’ll start with that before giving a couple other examples.

Odds of Romney winning

Taking yesterday’s New York Times 538 forecast chance of Mitt Romney winning each state, I entered the numbers into the search correlation machine. As you can see from the map on the left, these are very polarized numbers, with 42 of the states being above 90% or below 10%. Of the 100 Google searches whose relative frequency is most correlated with this pattern across states, 31 are about guns. Here is the search most correlated (.82) with Romney’s odds of winning: “marlin 30-30,” which is a classic rifle (available at Walmart):

Unintentional deaths

News the other day was about the lives lost to unintentional injuries for people under age 20 — the most common causes of death in that age range. About half of this is from motor vehicle accidents, with most of the rest distributed between drowning, suffocating, fires, falls, and poisoning. The CDC put out a report that included a state breakdown, reported in terms of “years of potential life lost” per 100,000 population. That is just the number of deaths times the number of years between the age at which the death occurred and age 75 (so a death at age 1 is 74 years lost, a death at age 19 is 56).

The big inequalities here are in gender and geography. Males are about 1.8-times more likely to die from this stuff. And the most dangerous state (Mississippi) has more than 4-times the losses of the safest (Massachusetts). There are race differences as well — with American Indians having high rates — but the Black/White difference is not that large (Latino ethnicity wasn’t identified).

How are these rates of lost life correlated with search behavior? Guns. Among the 100 searches that most closely follow the pattern of deaths, 62 were about guns, starting with number 1: “shotgun for sale,” with a correlation of .93.

There were also 14 searches about cars and trucks on the list (mostly Ford F150s and Chevy trucks), four about wedding dresses and rings (“discount wedding dresses”) and three about Fox News.


I did this twice with divorce rates. Using the 2008-2009 divorce rates per 1,000 married women, I found a good gun correlation with gun searches, with “colt .45 automatic” scoring a .86:

There were 27 more gun-related searches on that 08-09 divorce-correlation list. I updated that for the new 2011 rates, and again came up with a list of gun-related searches (and other military or survivalist stuff). Here is the Norinco SKS and 2011 divorce rates, correlation .84:

Someone who knows more than me could probably read more into the searches for different kinds of guns and gun-related stuff — for example, the difference between sniper accessories, shotguns and handguns. These different gun results show variations in their geographic patterns.

Anyway, I can’t think of what else besides search data tells us so much about so many people’s behavior — not their stated interests, their reported behavior, their tax forms, or their consumption patterns. And yet I can’t really put my finger on what it does tell us. It’s a million miles wide and not that deep, but it’s endlessly fascinating. If someone can figure out how to explain the value of what this all shows, I’m all ears.


Filed under In the news

Which comes first, getting divorced or Googling “vasectomy reversal”?

Quick follow up to the last post on the new 2011 divorce data.

I just noticed that the Census Bureau’s American Community Survey release included a calculation of the divorce rate, calculated as the number of divorces in the last 12 months per 1,000 people age 15+, with a table listing them by state (Factfinder tables GCT1253 and GCT1254).

They did it for men and women, and the rates are a little different, so I averaged them. Then, the first step in the research process is to run the state rates through Google Correlate

Of the top 100 searches most correlated with divorce rates — that is, searches that are most common where there are more divorces, and least common where there are few divorces — about 99 of them are about guns, military paraphernalia, survivalist stuff, etc.

And then, right between “radio software” and “shotgun pistol” there is “vasectomy reversal.” I am not making this up:

That’s the divorce rate on the left, and the “vasectomy reversal” searches on the right. It’s a correlation of .79 — pretty good.

Here are the other things on the list, all correlated with the divorce rate at between .77 and .84.


Filed under In the news, Me @ work

Where did we go wrong, or did we? (love and sex edition)

For ever – or at least for 70 years - make love was many times more common than have sex, at least in the Google Ngrams database of millions of books in American English. And, then — well, you can guess what happened then:

The results are the same with “making” and “having” (you can play with the search here).

Why? What happened? Could it be “the culture”? Zooming in on the period since 1950, preliminary evidence is mixed:

I’m open to hypotheses.




Filed under Me @ work

Strat theorists, ngram waves

As followup to today’s stratification syllabus, here’s how our leading theorists (at least the ones in my class) have fared in American English book references since the 1860s.

Two things. First is the waves. Marx, Weber and Parsons have their peaks in the early 1970s; Durkheim and MacKinnon peaked in the 1990s; Bourdieu may not have peaked yet as of 2008 (why doesn’t Google update this thing?).

Second is something about generational ripples, with within-theorist peaks repeating at intervals, such as Weber, Durkheim and (to a lesser degree) Parsons in 1970s and 1990s. The 70s and 90s might just be peaks in sociology publishing.

Anyway, the generational wave theory bodes well for a MacKinnon rebound.

You can play with this here.


Filed under Me @ work

A lot of juggling metaphors in the air

But what are people juggling?

[Updated with a look at phrase origins at the end.]

Judging by the prevalence of terms in the Google Ngrams database of books, since 1970 people have begun juggling their families, their work-and-family, their responsibilities, and even their children themselves.

Maybe all the metaphorical juggling has contributed to the rise of real-life juggling, although based on the distribution of juggling conventions this is a bigger deal in Europe than the U.S. In American English, “juggling balls” is thriving, but since the mid-1980s its growth can’t keep up with “juggling work.”

And, judging by who’s juggling in a Google image search for “juggling work,” the clip-artists of today, at least, think it’s women who are driving the trend about 2-to-1:

Origins update:

The Oxford English Dictionary doesn’t have a clear dating of this kind of use for juggling before 1985, “They have to know how to do many things—from juggling the futures market to overhauling a tractor or curing viral scours.”

The first instance of “juggling work” I get in the Lexis database is from the New York Times, Sep. 23, 1980:

She conceded that juggling her work and family is not always easy. ”You feel split all the time,” said Mrs. Massie, taking a cigarette. ”Sometimes the family responsibility collides with the need to be alone, and with the selfishness that is necessary for any creative effort.”

The American Sociological Review has a reference to “juggling work assignments” in a 1956 book review on industrial practices, which isn’t quite the sense of juggling tasks within a single life. By 1983 Frontiers: A Journal of Women Studies has this:

The impossible pressures of juggling work and family responsibilities have led some Soviet women to reject the ideology of emancipation altogether.

And by then we’re off and juggling.

Leave a comment

Filed under Me @ work

Recession divorce paper preview

8:30 AM, Thursday, May 3: Be at the first session of the Population Association of American conference to hear me present, “Divorce and the Recession, 2008-2010″ (and three other interesting papers on the how the recession has affected families).

The presentation version of my paper is now available as a Maryland Population Research Center Working Paper. Here are a few highlights and additions.

This analysis supersedes some of my earlier musings about divorce fluctuations, which have been quite inconsistent (here’s the whole series). I once reported a positive relationship between rising unemployment and rising divorce rates — but no increase in Google searches on divorce. But then my Google method turned up what looked like an increase in divorce-related searching – by which point I was skeptical that there was in fact dominant effect of the recession that is discernible in the short run. And now I don’t see an unemployment pattern to speak of.

The conference paper is the most I can do with what we now have — big-sample data from 2008-2010. As I noted before, there is a drop in divorce rates from 2008 to 2010, but that hides a rebound from 2009 to 2010; that pattern holds when individuals factors are controlled. In the context of a long-run decline in divorce rates, I don’t make much of that. At the state level, this my story:

After establishing an individual level model predicting women’s divorce, I test whether unemployment and foreclosures are associated with the odds of divorce, and for whom. Results show that foreclosure rates are positively associated with the odds of divorce, but only for those with more than a high school education. State unemployment rates show no effect on odds of divorce. I also test the effect of state laws delaying divorce, and find they have an increasingly negative effect of the three-year period, suggesting a backlog of new divorces during the recession.

The interpretation of those state law patterns — a late addition to the paper — is up for discussion. Anyway, here’s the figure showing the foreclosure pattern by education level, from a model that controls for individual characteristics and state fixed effects:

Maybe this means marriages in which people are more likely to own homes are more at risk of real estate shocks, but that’s pretty indirect. There might be a fancy way to work that out, with a prediction model for which of these divorced people probably owned a home before divorce (be my guest!).

Those state patterns are built on an individual model shown in this figure. Bars that point left show negative effects on divorce odds, bars that point right are for increased risks.

None of these patterns are surprising given past research, but it’s very nice to have recent big-data estimates as new benchmarks.

Finally, I updated the Google analysis, because I couldn’t resist. The trend for a basic “divorce” search, which I used previously, was seriously diverted by the something called “the Kardashian event” in October 2011. How much did this mess up the data? This much:

Partly for that reason, this time I stuck with lawyer searches: “divorce lawyer,” “divorce attorney,” and “family law attorney,” which are all pretty well correlated over time. This is the trend (dates on the x-axis appear at the end of each year):

I could interpret this as consistent with the divorce/recession lull-rebound hypothesis, but time will tell. It doesn’t fit well from 2004 to 2008, since divorce rates were probably falling during most of that time. Still, that’s a pretty rapid rise at the end. If there isn’t an increase in divorce in 2011/2012, remind me to report that this method didn’t work.

Leave a comment

Filed under Me @ work, Research reports