Disclaimer. Don't rely on these old notes in lieu of reading the literature, but they can jog your memory. As a grad student long ago, my peers and I collaborated to write and exchange summaries of political science research. I posted them to a wiki-style website. "Wikisum" is now dead but archived here. I cannot vouch for these notes' accuracy, nor can I even say who wrote them. If you have more recent summaries to add to this collection, send them my way I guess. Sorry for the ads; they cover the costs of keeping this online.
Fiorina. 1981. Some problems in studying the effects of resource allocation in Congressional elections. AJPS 25:543-567.
Responding to Johannes and McAdams's (1981) claims that casework has no electoral benefit, Fiorina presents a powerful critique of both their specifications and operationalizations. He presents evidence quite to the contrary: Voters do value casework--even casework rendered to other people--and they reward incumbents for it. (Be aware, though, that the authors present an equally powerful critique of Fiorina's arguments; see McAdams and Johannes 1981).
Johannes and McAdams (J&M) used two strategies to make their claim. First, they presented district-level data to show that members of Congress (MCs) who do more casework aren't more likely to be elected. Second, the present individual-level survey data to show that citizens aren't much more likely to support MCs as a result of casework they receive. Fiorina responds to both arguments.
McAdams and Johannes (M&J) responded forcefully to Fiorina's critiques; their responses follow each of Fiorina's critiques in the summary below. In particular, they take offense (apparently) at Fiorina's (implicit) suggestion that they came into their analysis seeking to disprove the casework hypothesis. In M&J's conclusion, they argue that they were actually expecting the opposite of what they found.
Mayhew (1974) raised the puzzle of the vanishing marginals. Fiorina (1977) suggested that casework was to blame ("the bureaucracy did it"). Johannes and McAdams (1981) challenge this claim. This article is Fiorina's response. McAdams and Johannes (1981) then responded to Fiorina's response.
In this summary, J&M refers to Johannes and McAdams 1981 (the first article) and M&J refers to McAdams and Johannes 1981 (the response to Fiorina's reply). J&M, Fiorina, and M&J all appeared in the same issue of AJPS.
Fiorina: J&M's district-level analysis suffered from two general types of problem: misspecification and measurement errors. Fiorina does not attempt to present a new district-level analysis; his argument (summarized below) makes clear that such an analysis requires more data than he has at present.
Analogy: Previous literature has found that the more incumbents spend, the worse they do. Jacobson (1978) explained this puzzle: Incumbents who expect a tough fight spend more than incumbents who expect to win easily; thus, spending more is evidence of a tough fight, so it isn't surprising that increased spending correlates with increased odds of failure.
Similarly, it shouldn't seem strange that J&M found that MCs who do more casework are not more likely to win. (In fact, they found insignificant results in the opposite direction.) Expectations about the next race influence MCs' decisions about casework today. And these expectations are influenced by the margin of victory in my last race--which, in turn, was related to the amount of casework I did before that race. Theoretically, this chain could go back to the incumbent's first race for House, though Fiorina suggests that we probably don't need to go back much further than the previous election and the previous term's casework (and the current casework and the expected results of the upcoming election). Fiorina summarizes the relationship between casework and election returns in equation 5, p 548.
Because J&M used simple regression analysis, they could not detect such a relationship. Thus, it is not strange that they found what they did--their methodology would also lead us to conclude that campaign spending lowers the probability of reelction.
M&J do not contest Jacobson's (1978) argument, but argue that it is not an appropriate analogy. MCs do fundraising in the months immediately before an election with an eye toward the short-term electoral results. However, Fiorina (1981) cites Fenno (1978), suggesting that casework is part of a long-term strategy that helps MCs build up support "over a long time." Moreover, Fiorina notes Fenno's argument that MCs tend to choose a "home style" early on and stick with it. Thus, it appears unlikely that casework should vary according to short-term electoral expectations. And if this is true, M&J argue, then the J&M model was not misspecified.
J&M group all casework into a single category, refusing to recognize that different types of casework have different effects. (i.e. casework for the local school board or mayor might have more of an effect than casework for an individual voter).
They rehash their earlier (J&M) argument that "high level" cases (1) are very rare and (2) are not necessary in Fiorina's original (1977) suggestion that it's the ordinary ("low level") casework that builds electoral support.
They also respond to Fiorina's criticism that each district's education levels may interact with casework by pointing out that they had explicitly run this analysis (bottom of M&J p 588; also M&J note 3).
Fiorina: The two major problems with J&M's individual-level analysis is its definition of casework and its multicollinearity. Fiorina uses the same 1978 CPS (Michigan) data that J&M used to produce the opposite results after making his corrections: Casework does matter.
J&M should have used both information requests and help requests as indicators of the same thing: casework (see p 552 for details). He presents an example to illustrate why. Suppose a constituent has a question about his eligibility for a government program or seeks an explanation of some bureaucratic process. Once CPS respondent might call this a request for information, while another might call it a request for help on a problem. J&M called it casework only if the respondent reported asking for help on a problem, needlessly reducing the number of people who had requested casework.
They argue that seeking help and seeking information really are distinct, and people perceive the difference (p 589). They "find it odd" that Fiorina criticizes them for separating these two types of distinct casework when he earlier criticized them for not separating "high-level" and "low-level" casework (in the district-level analysis).
In a later regression, J&M included both information requests and help requests as separate variables (see previous). Not only is this multicollinear, but both are multicollinear with the "contact" controls in J&M's analysis (b/c if you request help, you're likely to also report receiving a letter, meeting with a MC's staff, getting a phone call, etc). This major multicollinearity problem stifles significance.
First, Fiorina has overstated the degree of multicollinearity. M&J's table 4 (p 593) shows that being contact by an MC's staff is not excessively correlated with receiving case assistance; in fact, most measures of the correlation show a correlation of around 0.3. Only Gamma--which is inflated when used to measure the correlation of dichotomous variables--shows a high correlation (0.82); and it is Gamma that Fiorina used to argue that the right-hand variables were highly correlated.
Second, the authors present a more general argument that omitting variables from the analysis would be worse than including the extra ones that they have. They seek to show that selectively eliminating one of three variables (casework, seniority, and contact) would render other variables significant (Table 3), suggesting that Fiorina selectively omitted variables in order to find his significant effects.
Though Fiorina blasts J&M for ignoring ripple effects, looking for direct effects was actually the "most likely" place to look for validation of the casework hypothesis. "In short, we did not look for 'ripple effects'; we looked for the original splash. Not having found the splash (having looked precisely where it ought to be), we felt justified in discounting ripple effects" (M&J p 595).
Research by the same authors
Research on similar subjects