Skip to content

Jim Campbell argues that Larry Bartels’s “Unequal Democracy” findings are not robust

A few years ago Larry Bartels presented this graph, a version of which latter appeared in his book Unequal Democracy:


Larry looked at the data in a number of ways, and the evidence seemed convincing that, at least in the short term, the Democrats were better than Republicans for the economy. This is consistent with Democrats’ general policies of lowering unemployment, as compared to Republicans lowering inflation, and, by comparing first-term to second-term presidents, he found that the result couldn’t simply be explained as a rebound or alternation pattern.

The question then arose, why have the Republicans won so many elections? Why aren’t the Democrats consistently dominating? Non-economic issues are part of the story, of course, but lots of evidence shows the economy to be a key concern for voters, so it’s still hard to see how, with a pattern such as shown above, the Republicans could keep winning.

Larry had some explanations, largely having to do with timing: under Democratic presidents the economy tended to improve at the beginning of the four-year term, while gains under Republicans tended to occur in years 3 and 4–just in time for the next campaign!

See here for further discussion (from five years ago) of Larry’s ideas from the perspective of the history of the past 60 years.

Enter Campbell

Jim Campbell recently wrote an article, to appear this week in The Forum (the link should become active once the issue is officially published) claiming that Bartels is all wrong–or, more precisely, that Bartels’s finding of systematic differences in performance between Democratic and Republican presidents is not robust and goes away when you control the economic performance leading in to a president’s term.

Here’s Campbell:

Previous estimates did not properly take into account the lagged effects of the economy. Once lagged economic effects are taken into account, party differences in economic performance are shown to be the effects of economic conditions inherited from the previous president and not the consequence of real policy differences. Specifically, the economy was in recession when Republican presidents became responsible for the economy in each of the four post-1948 transitions from Democratic to Republican presidents. This was not the case for the transitions from Republicans to Democrats. When economic conditions leading into a year are taken into account, there are no presidential party differences with respect to growth, unemployment, or income inequality.

For example, using the quarterly change in GDP measure, the economy was in free fall in Fall 2008 but in recovery during the third and fourth quarters of 2009, so this counts as Obama coming in with a strong economy. (Campbell emphasizes that he is following the lead of Bartels and counting a president’s effect on the economy to not begin until year 2.)

It’s tricky. Bartels’s claims are not robust to changes in specifications, but Campbell’s conclusions aren’t completely stable either. Campbell finds one thing if he controls for previous year’s GNP growth but something else if he controls only for GNP growth in the 3rd and 4th quarter of the previous year. This is not to say Campbell is wrong but just to say that any atheoretical attempt to throw in lags can result in difficulty in interpretation.

I’m curious what Doug Hibbs thinks about all this; I don’t know why, but to me Hibbs exudes an air of authority on this topic, and I’d be inclined to take his thoughts on these matters seriously.

What struck me the most about Campbell’s paper was ultimately how consistent its findings are with Bartels’s claims. This perhaps shouldn’t be a surprise, given that they’re working with the same data, but it did surprise me because their political conclusions are so different.

Here’s the quick summary, which (I think) both Bartels and Campbell would agree with:

– On average, the economy did a lot better under Democratic than Republican presidents in the first two years of the term.

– On average, the economy did slightly better under Republican than Democratic presidents in years 3 and 4.

These two facts are consistent with the Hibbs/Bartels story (Democrats tend to start off by expanding the economy and pay the price later, while Republicans are more likely to start off with some fiscal or monetary discipline) and also consistent with Campbell’s story (Democratic presidents tend to come into office when the economy is doing OK, and Republicans are typically only elected when there are problems).

But the two stories have different implications regarding the finding of Hibbs, Rosenstone, and others that economic performance in the last years of a presidential term predicts election outcomes. Under the Bartels story, voters are myopically chasing short-term trends, whereas in Campbell’s version, voters are correctly picking up on the second derivative (that is, the trend in the change of the GNP from beginning to end of the term).

Consider everyone’s favorite example: Reagan’s first term, when the economy collapsed and then boomed. The voters (including Larry Bartels!) returned Reagan by a landslide in 1984: were they suckers for following a short-term trend or were they savvy judges of the second derivative?

I don’t have any handy summary here–I don’t see a way to declare a winner in the debate–but I wanted to summarize what seem to me to be the key points of agreement and disagreement in these very different perspectives on the same data.

One way to get leverage on this would be to study elections for governor and state economies. Lots of complications there, but maybe enough data to distinguish between the reacting-to-recent-trends and reacting-to-the-second-derivative stories.

P.S. See below for comments by Campbell.


  1. Chris says:

    Of course, when you look at the relationship between control of Congress and economic growth, the Republicans perform better.

    And, of course, Congress has many more tools with which to influence the economy than does the President.

    I'm not sure either stat is particularly meaningful. If we were doing this properly, there would be negative 30 or so degrees of freedom.

  2. Jim Campbell says:

    I think that the key to the difference between Bartels’ interpretation and mine is in the lagged effects of the economy. Personally, I cannot believe that the economy at time 1 is unrelated to the economy at time 2. It is just plain implausible. Does everything start from scratch when the year turns? That would be bizarre. Yet, using the one year lag either in GNP (note 37 in Bartels) or in income growth in particular percentiles (table 2.5 in Bartels), Bartels finds no significant positive effect of the economy in year 1 affecting the economy in year 2. Either we are to believe the implausible or we are to believe that the lag times are incorrect. I explored the later. I think that theory and common sense would strongly suggest that there should be a lagged effect, but would be agnostic over the particular length of the lag. Is it three months or six months or nine months? That is an empirical matter. I explored a one quarter lag in the economy and a half year lag in the economy. Using either lag, the presidential difference evaporates– for growth, for unemployment, and for income inequality. Are the economic lags perfect? No, but they are positive and significant as reason would suggest and this was not the case in Bartels' analysis.

    Moreover, since quarterly data are available for GNP growth, I explored shortening the presidential responsibility lag and using a one quarter lag in the economy. The presidential party difference again was not statistically significant.

    I think that two points are beyond contention (Hibbs or no Hibbs)-(1.) that the economy has a significant and positive lagged effect on the economic conditions that follow and (2.) that the economies inherited by each new Republican president since 1948 went into recession within months of the new president taking office (and that this was not the case for new Democratic presidents). Combine the two and it seems inescapable that the party differences in economic records were the result of inherited economic conditions and not differences in the success of the parties’ macroeconomic policies.

  3. David Weakliem says:

    I happened to have unemployment data from 1948 to 2010 on my computer, so I did a few regressions and found Bartels's conclusions were robust. If party in the White House is the only independent variable, the t-ratio is 3.3. If you add unemployment three and six months ago, the t-ratio for party is still 2.6 and the implied "equilibrium" difference in unemployment rates is large (about 2 percent). There's a hint that the Democratic advantage is bigger in the first years of the term, but even with that, predicted unemployment is always lower under the Democrats. So my analysis seems to support Andrew's judgment: "it's tricky."

  4. SH says:

    Jim, you can't just dump lagged DV's into the model without thinking about what you're doing.

    In most cases, the President's party was the same in the lagged period as it is in the current period. These lagged economic indicators could be outcomes influenced by the President's party in which case you're inducing post-treatment bias. In all likelihood, this will bias your estimated coefficient on "Democratic President" toward zero because the lagged economy is strongly, positively correlated with the current economy.

    In short, even though lagged economic indicators are strong predictors of the current economy (higher r-squared; lower standard errors, etc.) these are bad controls because they could plausibly be outcomes of the explanatory variable of interest. I'm surprised that Andrew didn't talk about this in his original post.

  5. numeric says:

    I'm curious what Doug Hibbs thinks about all this; I don't know why, but to me Hibbs exudes an air of authority on this topic, and I'd be inclined to take his thoughts on these matters seriously.

    You know, you were hammering Scott Adams for admiring Martin Sheen but your unqualified admiration of Hibbs is becoming disturbing. From the Harvard Crimson

    "And in 1985, Professor of Government Douglas A. Hibbs Jr. said he would resign after being accused of sexual misconduct, according to Crimson articles at the time" (

    The word in academy at the time was that this "misconduct" was forcible but Hibbs is certainly encouraged to waive confidentiality and have the records released to clear his name, inasmuch as an individual resigning from the top University in the world can clear his name.

  6. Andrew Gelman says:


    I was saying that Hibbs is an authority on elections and the economy. I don't think that describing someone as a subject-matter expert is the same as unqualified admiration.

  7. Steve Roth says:


    "Of course, when you look at the relationship between control of Congress and economic growth, the Republicans perform better.

    And, of course, Congress has many more tools with which to influence the economy than does the President."

    I don't know where those "of courses" come from. Parsing the data since '61, it seems obvious that the president's party is far more impactful than control of congress. At least the performance variances between pub and dem presidents are generally far greater than those between pub and dem senates and houses.

    As for economic performance:

    1. Democratic senates are associated with almost identical (barely lower) growth in GDP/capita compared to Republican ones.

    2. Democratic houses are associated notably better growth in GDP/capita.

    As for debt, Democratic houses are associated with much higher debt than republicans, and Democratic Senates are associated with much lower. Neither holds a candle to the difference between presidents, though. Republicans are associated with a 3.83% annual debt increase, while Democrats are associated with a 2.56% decrease.

    Nothing about this gives any impression of your statements being "obvious."

    Debt changes represent an average of 1-, 2-, 3, and 4-year lags. GDP/cap changes represent an average of 3- and 4-year lags (under the assumption that it's a second-order effect which takes time to happen.


  8. Paul E says:

    Douglas Hibbs is rather outdated these days. More up-to-date work on partisan business cycles was done by Alesina and also see Faust & Irons in the 1999 journal of monetary economics. They find no support for any partisan effects.

  9. 2slugbaits says:

    It's also revealing in terms of the implied economic theories. Republicans tend to support Real Business Cycle (RBC) economists, and RBC theory assumes the economy follows a stochastic trend (difference stationary) and that shocks have a permanent effect. Democrats tend to follow New Keynesian thinking, which usually assumes the economy is trend stationary and mean reverting but prone to periods of disequilibrium. Democrats explain unemployment as a level shift with government's job being that of maintaining aggregate demand until the economy reverts to its long run growth trend. Republican RBC-style economists (e.g., John Cochrane) tend to interpret unemployment as a supply side shock do the some exogenous decision by workers to voluntarily seek more leisure. In their view the role of GOP policies should be to induce a different labor/leisure trade-off by lowering taxes.

    So I suppose it's no surprise that Democratic leaning analysts would emphasize policies that tend to be more long-term oriented because they believe policies should be designed towards mean reversion; meanwhile, Republican leaning analysts tend to emphasize the need for differencing and/or longer lags in order to understand the dynamics.

  10. Jim Campbell says:

    Thanks for your comments. In response to “you can't just dump lagged DV's into the model without thinking about what you're doing.” These were not mindlessly dumped lagged DV’s. There are strong reasons to expect prior economic conditions to affect later economic conditions. Indeed, a model without significant lagged effects of the economy is highly suspect. Bartels implicitly concurred with this since he tested for lagged economic and income effects, only not of the right duration.

    Second, in response to the possibility that the prior economy is an outcome of party differences (of some unknown cause since this is a black box in the original study), the correlation of the presidential party is not strongly correlated with the lagged economy (correlations of about .25 with the 3rd or 4th quarter lags). Additionally, just as an initial examination of the possibility you suggest, there are no significant party differences in the lagged 4th quarter growth when controlling for 3rd quarter growth.

    Now I could keep pushing back for a party difference in lagged economic effects, but the major points remain that (1.) there are lagged effects of economic conditions on later economic conditions (and that these were not taken into account adequately in the original Unequal Democracy analysis), and that (2.) the economies were going into recession within months of each of the four new Republican presidents taking office since 1948 and that this was not the case in any of the transitions for new Democratic presidents. Four for four versus zero for four.

    Since everybody seems to agree that party differences are largely located in the second year of presidential terms, it would seem that we should give greater credence to hard evidence linking the second year differences to inherited recessions in the first year than to what are essentially speculations about differences in the success of party policies in honeymoon years. The basis for the speculation seemed to be that other explanations (including inherited economies) had been ruled out and so policy differences, though unexplored, must be the answer. My article indicates not only should the inherited economy explanation not have been ruled out, but that there is strong direct evidence supporting the inherited economy explanation (unlike the policy difference explanation).