Dependent Variable

Following Kung and Chen (Reference Kung and Chen2011), I use excessive grain procurement—which they define as the difference between normal levels of procurement during 1955–57 and actual procurement levels during 1958–65—as a proxy for political radicalism.Footnote ⁷ Grain procurement is chosen to proxy for political radicalism because, in a context in which Mao intended to industrialize China rapidly by moving as much agricultural surplus to the urban sector as possible, it unambiguously signaled commitment to the Great Leap, and because its variations across provinces significantly affected death tolls. Because I want to exploit the variation in the ballot outcome of the 8th National Congress (which first took place in September 1956) to examine the effect of membership status on political radicalism using the RD design,Footnote ⁸ I extend the study period for actual procurement levels to include the year 1957. Doing so requires me to correspondingly extend the study period for normal levels of procurement by one year to 1954–56. To also control for the period after the Great Leap, I choose the year 1965 as the final year of analysis (instead of 1966, as Kung and Chen (Reference Kung and Chen2011) do). I drop 1966 from analysis on the grounds that the Cultural Revolution, which broke out in May 1966, may contaminate our analysis. I plot the procurement ratio, average procurement ratio and excessive procurement ratio in Figure 1, which shows that grain procurement during the 1959–61 period was indeed excessive compared to the post-famine years.

Fig. 1

Grain procurement ratios, by year

Key Independent Variables

My main goal is to test whether AMs were more radical politically than FMs. The key independent variable of interest is therefore membership status, which is defined according to whether a provincial first party secretary was an FM, an AM or a nonmember (NM), using FMs as the reference group in the analysis.Footnote ⁹ As complications may arise if two party secretaries were found to serve in the same province in a given year, I employ the criterion adopted by Li and Zhou (Reference Li and Zhou2005) and select the party secretary who served on 30 June of that year as the representative leader of that province, and use her membership status accordingly. For instance, in Liaoning Province, Huang Oudong ended his tenure as the province's first party secretary in June of 1958, with Huang Huoqing replacing him immediately in the same month. While the precise date of this turnover is not specified, what is clear is that Huang Huoqing was in charge of Liaoning Province on June 30. Going by our rule, I choose Huang Huoqing (rather than his predecessor) to be the first party secretary of Liaoning Province for the year 1958.Footnote ¹⁰

Individual Characteristics of the First Party Secretary

Here I control for the age, years of party membership and “revolutionary credentials” (experience of the Long March is the pertinent proxy) of the provincial first party secretaries.Footnote ¹¹ In the extension using the RD design, I also control for their education (as a measure of competence) and add a new variable, Anti-Japanese Military and Political University (AJMPU, or “Resistance University”), to check the robustness of their Long March credentials.Footnote ¹² Since the AJMPU was the training ground for both military and political cadres of the CCP, I consider a province's first party secretary who was affiliated with the AJMPU—in the capacity of an administrator or instructor (or even a student)—as having significantly greater (revolutionary) credentials than those without such affiliations, and I construct a dummy variable to reflect that difference.Footnote ¹³ In addition to being a corroborating proxy, this broader categorization of the AJMPU should also permit a more encompassing definition of revolutionary credentials than the Long March.Footnote ¹⁴

Provincial Socio-economic Variables

Going beyond the individual level, I control for the possible effects of food availability, natural calamities, share of agriculture and GDP per capita. While I define these variables in basically the same way as Kung and Chen (Reference Kung and Chen2011) do, to avoid reverse causality I lag these economic variables by one year, as excessive procurement in a given year may affect these socio-economic variables in the subsequent year. As for Party Membership Density (a proxy for local variations), instead of defining it as the ratio of the number of party members to the total number of cadres in a province (Kung and Chen Reference Kung and Chen2011), I change the denominator to a province's population, as there are simply too many missing observations based on the previous definition. Table 1 defines all the variables and data sources (see Appendix Table A1 for details of data construction).

Table 1

Variables Definition and Data Sources

(1) N: observations; n: number of provinces; (2) Sources. A: Materials on the Agricultural Economy, 1949–83; B: Comprehensive statistical data and materials on 50 years of new China; C: Comprehensive statistical data and materials on 60 years of new China; D: A Compendium of Central Committee Members of Various Plenums, 1921–2003; E: Shih, Shan and Liu (2010); F: Report on China's Natural Disasters; G: Materials on the Organizational History of China's Central Communist Party. *Kung and Chen (Reference Kung and Chen2011).

A Congressional Ballot Outcome and CC Membership Ranking

Just as Thistlethwaite and Campbell (Reference Thistlethwaite and Campbell1960) analyze the impact of merit awards on future academic outcomes, here I study the possible effect of CC membership status on radical political behavior—specifically excessive grain procurement. Following their central idea—that individuals who ranked just below a certain cutoff point for an award (that is, those who just missed out on an award) serve as good comparisons for those who ranked just above the cutoff value (that is, those who just exceeded the threshold for an award)—I assume that the AMs who ranked near the cutoff point for full CC membership status were good comparisons to the FMs who also ranked near that cutoff point. To prove that this context satisfies the RD requirement that the individuals concerned were unable to precisely sort themselves around the discontinuity threshold (that is, to assign a particular membership status to themselves, say FM), I first provide a descriptive analysis of the pertinent institutional context, specifically the mechanisms for generating membership rankings. I then explain the paradox of why, in spite of this rigid institutional feature, the AMs were more radicalized by their desperate desire for promotion than the FMs, who could not hope to advance much further.

Although beginning in around the late 1950s Mao was increasingly behaving like a dictator,Footnote ¹⁸ the CCP had already developed formal political institutions governing membership, protocol, etc., as well as a set of formal rules and procedures designed to “both facilitate the exchange of information among the ruling elites and provide for an easy assessment of compliance with those rules” (Svolik Reference Svolik2012, 7; see also MacFarquhar Reference MacFarquhar1997). Given the political importance of the CC-CCP, there was an elaborate procedure for: (1) governing membership eligibility; (2) determining, by voting, the relative rankings of those who are eligible; and (3) specifying a precise cutoff point that effectively divides the CC members into two classes—FMs and AMs. These procedures made it virtually infeasible for the eligible nominees to manipulate the voting to the extent of influencing the final ranking, which determined membership status.

In a nutshell, a total of 1,026 delegates competed for one of 170 seats on the CC.Footnote ¹⁹ The 170 finalists were chosen only after three rounds of voting. After the first round of voting, a revised list was presented to the delegates, who were then asked to conduct another round of voting. Those voted on in the second round were then presented to the delegates again for a final secret ballot, in which each of the 1,026 delegates was given a form that contained the names of the 170 nominees who made it through the first two rounds. The pertinent instruction, for these delegates, was to strike out the names of those they did not support for a place in the CC. Unless a nominee was rejected by all 1,026 delegates, his place on the CC was now assured and his ranking determined based on the outcome of this third and final round.Footnote ²⁰

As each eligible nominee could be voted on more than once, and although the 170 finalists would most certainly have lobbied the Congress delegates to vote for them, it is unlikely that they were able to precisely influence the number of votes they each received to the extent of affecting their eventual CC ranking (and accordingly, their membership status). In fact, with the sole exception of Mao, even the top leadership would have found it a formidable task to control the number of votes to the extent of tallying the ballot outcome with that of their own preference.Footnote ²¹ At least that appeared to be the case for the Politburo. For example, with the exception of Mao Zedong and Liu Shaoqi, the “internal” rankings of the other 15 Politburo members all differed from the ballot outcome (Appendix Table A5). This particular mechanism of determining CC membership ranking (and, accordingly, the assignment of members to either FM or AM status) thus prevents individual CC members from precisely sorting themselves around the discontinuity threshold; this feature importantly qualifies the analysis for an RD design.Footnote ²²

Another distinctive feature of the mechanism of determining membership ranking is that, once the members’ rankings were determined by the final ballot outcome, future changes in ranking became automatic: one would automatically move up the rank upon the death of a more senior member, and would be promoted to full membership after crossing the threshold that separates FMs from AMs. A CC member would thus become an FM upon coming 97th (or higher) in the rankings. This was the case, for example, of Wang Enmao, the first party secretary of Xinjiang Province. Ranked 99th on 30 June 1957, Wang moved up two notches to rank 97th a year later and was promoted to FM, thanks to the deaths of two FMs.Footnote ²³ Similarly, Ye Fe, the first party secretary of Fujian Province (who ranked 103rd in 1958) advanced six notches by 1966 to rank 97th upon the demise of six FMs during this period and was thus promoted to full membership that year.Footnote ²⁴ The ballot results for all CC members who held the first party secretary position between 1956 and 1966 are summarized in Appendix Table A6 (and presented visually in Figure 2), which shows that a CC member indeed qualified for full membership upon coming 97th in the rankings. In short, this automatic feature of membership movement prevents individuals from rigging the votes and affecting the ranking accordingly.

Fig. 2

The relationship between ballot ranking and membership status

Figure 2 previews the discontinuity in membership status by plotting the bivariate relationship between Membership for the 1957–66 period and Ballot Ranking—our assignment variable. The fitted values, which are obtained from nonparametric regressions using Cleveland's (Reference Cleveland1979) tricube weighting function and a bandwidth of 0.5, are estimated separately for observations on either side of the cutoff value. The fitted values may thus be considered to represent the moving average of membership status across the entire range of ballot rankings; the scattered data points (hollow squares and circles) represent the membership of individual CC members. Figure 2 clearly shows that there is a discontinuity in membership status at the thresholds.

If advancement through the ranks was mechanically determined by what was essentially the natural attrition of the FMs, does it not contradict our central thesis regarding the differential incentives between AMs and FMs whose rank was close to the discontinuity threshold? The answer is no. The differential incentives between the two membership statuses stem from the fundamental fact that the probability that an AM will acquire the higher status of FM at the next congressional meeting is smaller than the probability that an FM will retain the same status. This difference thus provided sharp enough incentives for the AMs to exert greater efforts during the Leap in the hope of gaining more votes at the next congressional ballot. Moreover, the fact that nobody knew just when the next congressional session would take place helped create the expectation (illusion) that promotion could come sooner than the “constitutional” rule dictated.Footnote ²⁵

Mao's Influences and Possible Omitted Variable Bias

While Kung and Chen (Reference Kung and Chen2011) controlled for the omitted variable of personal identities in their fixed-effects estimations, the “personalities” or political dispositions of provincial leaders may change over time in response to Mao's persuasions. As Mao became increasingly impatient with the slow pace of economic development (for which he blamed the cautiously “expert” approach of the planners),Footnote ²⁶ he took a number of carefully orchestrated steps to persuade the provincial leaders—the actors Mao relied upon the most to spur the Leap—to increase production targets. For instance, at a number of ad hoc conferences that Mao set up and enlisted the unprecedented participation of the provincial leaders,Footnote ²⁷ he attended only those sessions presided by the local leaders (Teiwes and Sun Reference Teiwes and Sun1999). Moreover, of the dozen or so provincial party secretaries who accompanied Mao during his campaign tours in 1958, half were FMs; as many as four of them ranked just above 97 in the CC and one was exceptionally inducted into the Politburo in the same year (see below).Footnote ²⁸

As long as some provincial leaders responded to this deliberate act of Mao in the way he intended, the fixed-effects estimation would fail to account for the effect of this possible influence—one that potentially affects both career incentives and political radicalism simultaneously. By exploiting the variation in CC membership ranking determined by congressional ballot—which is uncorrelated with personality but which can affect membership status, the RD design serves to identify the latter's causal effect on political radicalism, thereby correcting for the possible omitted variable bias (of “personality change”) arising from the provincial leaders’ interactions with, and their differential responses to, Mao in the fixed-effects model.

Model Specifications

Given that membership status (AM_it) is automatically determined (discontinuously) by a member's ranking (R_it), I employ the sharp RD design to exploit its effect on political radicalism, conditional upon ranking (R_it). Specifically, given that the discontinuous jump in the status of CC membership occurs at the cutoff point c of 97, we have

$$A{{M}_{it}}\, = \,\left\{ \matrix { {\rm{1}} & {{\rm{if}}\,{{R}_{it}} \gt c} \cr{\rm{0}} & {{\rm{if}}\,{{R}_{it}}\leq c} \\\end{}} \right..$$

To allow for an easier interpretation of the empirical results, I normalize ranking as a function of $$$ {{r}_{it}}\, = \,\frac{{{{R}_{it}}\,{\rm{ - }}\,c}}{c} $$$ , which allows the function of membership status to be written as:

$$A{{M}_{it}}\, = \,\left\{ \matrix {{\rm{1}} & {{\rm{if}}\,{{r}_{it}}\, \gt \,{\rm{0}}} \cr{\rm{0}} & {{\rm{if}}\,{{r}_{it}}\,\leq \,0} \\\end{}} \right.$$

and the RD estimates to be constructed as:

$$\matrix {{EP{{R}_{it}}} \hfill & {\, = \,A{{M}_{it}}EP{{R}_{it}}(1)\, + \,(1\,{\rm{ - }}\,A{{M}_{it}})EP{{R}_{it}}(0)} \hfill \cr\ {} \hfill & {\, = \,\left\{ \matrix {{EP{{R}_{it}}(1)} \cr{EP{{R}_{it}}(0)} \\\end{}} \right. = \left\{ \matrix{{\beta \cdot 1\, + \,{{f}_{\rm{1}}}({{r}_{it}})\, + \,{{X}_{it}}\gamma \, + \,{{Z}_{it}}\rho \, + \,{{\alpha }_i}\, + \,{{\lambda }_t}\, + \,{{\upsilon }_{it}}} & {{\rm{if}}\;{{r}_i}\, \gt \,{\rm{0}}} \cr{\beta \cdot 0\, + \,{{f}_{\rm{0}}}({{r}_{it}})\, + \,{{X}_{it}}\gamma \, + \,{{Z}_{it}}\rho \, + \,{{\alpha }_i}\, + \,{{\lambda }_t}\, + \,{{\upsilon }_{it}}} & {{\rm{if}}\;{{r}_i}\,\leq \,{\rm{0}}} \\\end{}} \right} \hfill \\\end{}}.$$

Assuming that f ₀ (r_it) and f ₁ (r_it) are continuous in the neighborhood of zero (a standard assumption of the RD design), we can also rewrite Equation 3 as:

$$EP{{R}_{it}}\, = \,\beta \cdot A{{M}_{it}}\, + \,{{f}_1}({{r}_{it}})\, + \,[{{f}_1}({{r}_{it}})\,{\rm{ - }}\,{{f}_0}({{r}_{it}})] \cdot A{{M}_{it}}\, + \,{{X}_{it}}\gamma \, + \,{{Z}_{it}}\rho \, + \,{{\alpha }_i}\, + \,{{\lambda }_t}\, + \,{{\upsilon }_{it}}.$$

Figure 3 visually previews the RD results by plotting the bivariate relationship between the degree of radicalism for the period 1957–65 and ballot ranking. As explained earlier, the fitted values are obtained from nonparametric regressions using Cleveland's (Reference Cleveland1979) tricube weighting function and a bandwidth of 0.5 estimated separately for observations on either side of the cutoff value.Footnote ²⁹ The fitted values thus represent the moving average of excessive grain procurement across the entire range of ballot rankings; the scattered data points represent the various levels of political radicalism of individual CC members. Figure 3 presents clear evidence that the AMs tended to procure more grain.Footnote ³⁰ An especially convincing feature of the graph is that it provides clear evidence of a discontinuous increase in grain procurement at the juncture r_i=0. This jump is meaningful, as it seems improbable for other determinants of grain procurement to change as suddenly at the cutoff point as membership status.Footnote ³¹

Fig. 3

The relationship between ballot ranking and excessive grain procurement ratio

Now turning to the estimations, although nonparametric techniques are advantageous in that they do not rely on the assumption of functional forms, the small size of our sample obligates us to employ parametric estimations instead. As Lee and Lemieux (Reference Lee and Lemieux2010) suggest, a simple way of relaxing the linearity assumption in parametric estimations when the true functional form is unknown is to “include polynomial functions of X in the regression model” (316). We thus model f ₀ (r_it) and f ₁ (r_it) as functions with varying degrees of polynomials, and rewrite Equation 3 as:

$$\matrix {{E(EP{{R}_{it}}|A{{M}_{it}}\, = \,0)} \hfill & {\, = \,{{\psi }_{01}}{{r}_i}\, + \,{{\psi }_{02}}r_{i}^{2} \, + \, \cdots \, + \,{{\psi }_{0p}}r_{i}^{p} \, + \,{{X}_{it}}\gamma \, + \,{{Z}_{it}}\rho \, + \,{{\alpha }_i}\, + \,{{\lambda }_t}\, + \,{{\upsilon }_{it}}} \hfill \cr{E(EP{{R}_{it}}|A{{M}_{it}}\, = \,1)} \hfill & {\, = \,{{\psi }_{11}}{{r}_i}\, + \,{{\psi }_{12}}r_{i}^{2} \, + \, \cdots \, + \,{{\psi }_{1p}}r_{i}^{p} \, + \,{{X}_{it}}\gamma \, + \,{{Z}_{it}}\rho \, + \,{{\alpha }_i}\, + \,{{\lambda }_t}\, + \,{{\upsilon }_{it}}} \hfill \\\end{}$$

and the regression model as:

$$EP{{R}_{it}} = \beta \cdot A{{M}_{it}}\, + \,\mathop{\sum}\limits_{j\, = \,{\rm{1}}}^P ({{\kappa }_{0j}}r_{{it}}^{j} \, + \,\kappa _{j}^{\ast} r_{{it}}^{j} A{{M}_{it}})\, + \,{{X}_{it}}\gamma \, + \,{{Z}_{it}}\rho \, + \,{{\alpha }_i}\, + \,{{\lambda }_t}\, + \,{{\upsilon }_{it}},$$

where $$${{\kappa }_{0j}} = {{\psi }_{0j}}$$$ and $$$\kappa _{j}^{\ast} \, = \,{{\psi }_{1j}}\,{\rm{ - }}\,{{\psi }_{0j}}$$$ . In the case of polynomial regressions, the key question concerns the choice of the order of the polynomial regressions, which can be decided using the Akaike information criterion (AIC) of model selection. In the RD context, AIC is given by $$$nln({{\mathop{\sigma }\limits^{\frown}}^2} ) + 2(P + 1)$$$ , where $$$\mathop{\sigma }\limits^{\frown}$$$ is the mean squared error of the regression and P is the number of parameters in the regression model (order of the polynomial) (Lee and Lemieux Reference Lee and Lemieux2010). The AIC can be minimized by choosing P.

Empirical Results

Baseline results

Model 1 of Table 3 (the upper panel) examines the robustness of the polynomial r_it by means of six specifications. Columns 1–6 report the results that include AM_it, the first through to the sixth-order polynomials of r_it, and the interactions between AM_it and the polynomials of various orders—linear (P=1), quadratic (P=2), cubic (P=3) and so on. We also report the AIC statistics for all the specifications. The optimal order of polynomial, which can be decided by minimizing the AIC, is four, implying that it is necessary to control for the quartic polynomial and its interactions with AM_it. That the fourth-order polynomial is optimal can easily be seen from the fact that the AIC that controls for the cubic polynomials and their interactions with AM_it (Column 3) is only trivially larger (by implication of the AIC) than the one controlling for a higher-order polynomial (Column 4), and that the estimated effects of AM_it are strikingly similar between these two specifications. Conversely, the models with lower-order polynomials are suboptimal because they result in significantly smaller estimates and larger AICs (Columns 1 and 2). Overall, the robustness of our estimations is borne out by the fact that AM_it continues to exhibit a highly significant and positive effect on the excessive grain procurement ratio even in estimations that include higher-order polynomials and their interactions with AM_it (Columns 5–6), suggesting that our results are not sensitive to the order of the polynomial. In terms of magnitude, the RD estimate yields a coefficient of 8.734 (Column 4, Table 3)—more than three times the estimate of the fixed-effects model (2.826, Column 7, Table 2). To verify that these results are not confounded by the individual characteristics of the provincial leaders, I dropped all the personal characteristics in Model 2 (Table 3), and found that the results changed only trivially (compare, for example, the coefficient of 8.986 with 8.734 in Column 4 of the two models)—the former is larger by a mere 0.25.

Table 3

Regression Discontinuity Estimations of Alternate Members’ Political Radicalism

Note: robust standard errors in parentheses. *significant at 10%; **significant at 5%; ***significant at 1%. Constant terms are included but not reported. Control variables include all control variables in Column 4 of Table 2.

Given that the RD results may be sensitive to the choice of bandwidth, I replicate our estimations using different bandwidths to ensure that our baseline results are robust.Footnote ³² I do so by using the alternative bandwidths of (−0.4,0.4), (−0.5,0.5), (−0.6,0.6), (−0.7,0.7) and (−0.8,0.8), based on the specification in Column 4 of Table 3 (Model 1). These checks show that the estimates do not change systematically with the change in bandwidth. In particular, compared with the estimates based on the whole sample (8.734 in Column 4 of Table 3), the magnitudes of the coefficient (of AM) fall within a remarkably close range of that using the original bandwidth (see Appendix Table A7 for details).

Robustness check

A concern arises, however, if the probability of receiving the treatment does not change from 0 to 1 at the threshold, in which case we need a design that allows for a smaller jump in the probability at the threshold (Imbens and Lemieux Reference Imbens and Lemieux2007). Table A6 indeed shows that “compliance” with the institutional rules and mechanisms is less than perfect—that is, while some AMs were already ranked 97th before 1966 (for example, Wei Guoqing in 1961 and Ye Fei in 1965), they did not officially become FMs until August 1966 (the 9th National Congress). This time lag before an individual was granted full membership implies that the treatment in question is less than 1 (r_i ≤ 0), which suggests that it is necessary to check the robustness of our previous estimates using the fuzzy RD design.Footnote ³³ We thus assume in this context that the probability of being an AM is smaller than 1, as specified in Equation 6:

$$P(A{{M}_{it}})\, = \,\left\{ \matrix {{\rm{1}} & {{\rm{if}}\,{{r}_{it}}\, \gt \,0} \cr F & {{\rm{if}}\,{{r}_{it}}\,\leq \,0} \\\end{}} \right..$$

We also assume that alternate membership can be written as:

$$E(A{{M}_{it}})\, = \,\phi \cdot {{D}_{it}}\, + \,\mathop{\sum}\limits_{j\, = \,{\rm{1}}}^P ({{\varphi }_{0j}}r_{{it}}^{j} \, + \,\varphi _{j}^{\ast} r_{{it}}^{j} {{D}_{it}})\, + \,{{X}_{it}}\rOmega \, + \,{{Z}_{it}}\theta \, + \,{{\alpha }_i}\, + \,{{\lambda }_t},$$

where D_it is a dummy variable indicating a ranking lower than the cutoff value, namely r_i > 0. The most straightforward method of estimating Equation 7 is to regress AM_it on D_it and a set of control variables first, followed by regressing the excessive procurement ratio on the predicted values of AM_it and the same set of control variables. The results are reported in Table 4. Specifically, AM_it remains significant after controlling for the cubic and quartic polynomials and their interactions with AM_it (Columns 1 and 2). Their magnitudes, which are bound between 7.911 and 7.985, are only slightly smaller than the baseline results of 8.532 and 8.734 (Columns 3 and 4 in Table 3). As Ye Fei and Wei Guoqing were not promoted until several years after they reached the threshold (see Table A6), to ensure that our results are not driven by these two special cases I dropped them from the analysis. Columns 3 and 4 of Table 4, which reports these alternative results, show that the effect of AM_it now becomes 8.638, which is strikingly similar to the pertinent coefficient of 8.734 in Table 3 (Column 4). The result of the fuzzy RD estimation thus confidently reaffirms the baseline results.

Table 4

Regression Discontinuity Estimations: Robustness Checks

Note: robust standard errors in parentheses. *significant at 10%; **significant at 5%; ***significant at 1%. Constant terms are included but not reported. Control variables include all control variables in Column 4 of Table 2.

The RD approach requires the identifying assumption that, except for the assignment variable, all outcomes must vary smoothly at the cutoff point (that is, they have to be continuous). Suppose W ¹ and W ⁰ represent the potential outcomes in the two groups with AM_it=1 and AM_it=0, respectively. Then E (W ¹ | x,Z) and E (W ⁰ | x,Z) must be continuous at the cutoff point. I evaluate this assumption in the Appendix, and find that AM_it is uncorrelated with any of the control variables including Age, Years of Party Membership, Education (a proxy for competence), and Long March Experience and AJMPU (proxies for revolutionary credentials) (see Table A8). Moreover, even with the inclusion of Education, Long March Experience and AJMPU, the effect of AM_it remains highly significant (see Appendix Table A9).Footnote ³⁴

Placebo test

Although the above robustness check confirms that membership status has no significant effect on W_i (comprising age, years of party membership, and education and revolutionary credentials), and that it remains significant even after controlling for these variables, there may still be other unobserved changes around the cutoff value. To rule out this possibility, I perform a placebo test by changing the discontinuity threshold.Footnote ³⁵ If doing so does not alter our previous results, we can be more confident of our RD estimates. We assume, for instance, that the discontinuous jump in the status of CC membership occurs at points other than 97. We then have

$${{C}_{it}}\, = \,\left\{ \matrix {{\rm{1}} & {{\rm{if}}\,{{R}_{it}}\, \gt \,c} \cr{\rm{0}} & {{\rm{if}}\,{{R}_{it}}\,\leq \,c} \\\end{}} \right.,$$

in which I assume that c=90,93,…96 and 98,99, …104. Similarly, I normalize ranking as a function of $$${{r}_{it}}\, = \,\frac{{{{R}_{it}}\,{\rm{ - }}\,c}}{c}$$$ , which allows the function of membership status to be written as

$${{C}_{it}}\, = \,\left\{ \matrix {{\rm{1}} & {{\rm{if}}\,{{r}_{it}}\, \gt \,{\rm{0}}} \cr{\rm{0}} & {{\rm{if}}\,{{r}_{it}}\,\leq \,{\rm{0}}} \\\end{}} \right.$$

and the RD estimates to be constructed as

$$EP{{R}_{it}}\, = \,\beta \cdot {{C}_{it}}\, + \,\mathop{\sum}\limits_{j\, = \,1}^P ({{\kappa }_{0j}}r_{{it}}^{j} \, + \,\kappa _{j}^{\ast} r_{{it}}^{j} {{C}_{it}})\, + \,{{X}_{it}}\gamma \, + \,{{Z}_{it}}\rho \, + \,{{\alpha }_i}\, + \,{{\lambda }_t}\, + \,{{\upsilon }_{it}}.$$

I plot the estimates of β in Figure 4 along the x-axis according to different thresholds. The results show that the effect of other cutoff values is not significant: only discontinuity at the cutoff point of 97 has a significant effect on grain procurement, which suggests that the RD approach does capture the effect of the membership assignment rule instead of other unobserved changes. To check the robustness of this result, I further assume that c takes on other, more extreme, values—specifically 60, 75, 119 and 134. The results show that the effects of these arbitrary cutoff values are not significant. Moreover, for cutoff values that are substantially smaller than 97 (for example, 60), the standard deviation becomes considerably larger due to the drastically reduced number of observations around this cutoff value.

Fig. 4

Placebo test: effect of changing the discontinuity threshold

The Discontinuous Jump in Political Radicalism between AMs and FMs

A unique feature of the RD design is that, instead of measuring the average effect of the explanatory variable on the dependent variable, it measures the marginal effects, because RD is concerned with the effects of only those values lying around the discontinuous threshold. This can be illustrated by Equation 9, which is rewritten from Equation 5:

$$\matrix { {EP{{R}_{it}}} \hfill & {\, = \,\overline{\beta } \cdot A{{M}_{it}}\, + \,\mathop{\sum}\limits_{j\, = \,1}^P {{\kappa }_{0j}}r_{{it}}^{j} \, + \,{{X}_{it}}\gamma \, + \,{{Z}_{it}}\rho \, + \,{{\alpha }_i}\, + \,{{\lambda }_t}\, + \,{{\upsilon }_{it}}} \hfill \cr{} \hfill & {\, = \,\left( {\beta \, + \,\mathop{\sum}\limits_{j\, = \,1}^P \kappa _{j}^{\ast} r_{{it}}^{j} } \right)A{{M}_{it}}\, + \,\mathop{\sum}\limits_{j\, = \,1}^P {{\kappa }_{0j}}r_{{it}}^{j} \, + \,{{X}_{it}}\gamma \, + \,{{Z}_{it}}\rho \, + \,{{\alpha }_i}\, + \,{{\lambda }_t}\, + \,{{\upsilon }_{it}}} \hfill \\\end{}}.$$

While the fixed-effects model (first line in Equation 9) estimates $$$\bar{\beta }$$$ without controlling for any polynomials of the assignment variable,Footnote ³⁶ the RD measures, in this instance, the effect of membership status (allocated on the basis of ballot ranking) around the threshold that separates the AMs from the FMs, β, on excessive grain procurement (second line in Equation 9). This explains why the magnitude of AM_it, according to the RD estimate (8.734, Column 4, Table 3), is so much larger than the fixed-effects estimate (2.826, Column 7, Table 2).

To illustrate this unique property of the RD design in our context of interest, I plot the coefficients of the term $$$\bigg(\big(\beta \, + \,\mathop{\sum}\nolimits_{j\, = \,1}^P \rho \kappa _{j}^{\ast} r_{{it}}^{j} \big)A{{M}_{it}} + \mathop{\sum}\nolimits_{j\, = \,1}^P \kappa _{{0j}}^{\ast} r_{{it}}^{j} \bigg)$$$ in Figure 5 based on the results in Table 3 (Column 4). The x-axis measures the ballot ranking of the CC members using the normalized term $$$\frac{{{{R}_{it}}\,{\rm{ - }}\,c}}{c}$$$ , whereas the y-axis measures the degree of radicalism. The solid, vertical line in the middle represents the cutoff point that separates AMs and FMs; specifically, it represents the degree of radicalism of the AMs to the right and the degree of radicalism of the FMs to the left. A striking feature of Figure 5 is that political radicalism experiences a discontinuous jump as the ballot ranking turns from positive (to the right of the vertical line) to negative (to the left)—a feature that confirms that the difference in the degree of political radicalism between FMs and AMs is driven by the difference in membership status.Footnote ³⁷

Fig. 5

Degree of political radicalism, by ballot ranking

Specifically, Figure 5 reveals that, in the case of AMs, those who came closest to the cutoff point (of 97) are indeed the most radical. Being on the verge of promotion, these CC members hardly needed any persuasion from Mao to toe his line. But radicalism declines in tandem with the decrease in ranking (reflected in the increase in ballot ranking), suggesting that those with significantly lower rankings saw little hope of promotion. In sharp contrast to the highest-ranked AMs, FMs who only just made it are the least radical, for they must have anticipated only a small likelihood of promotion to the Politburo and so had distinctly weaker incentives to procure more grain. Consistent with the logic discussed earlier, however, radicalism rises in tandem with ranking. These patterns of relationship are highly robust: the shapes of the curves are strikingly similar regardless of the order of the polynomials (cubic and quartic) used.

Countering Political Complacency: Mao Strikes Back

Our findings so far are seemingly inconsistent with the fact that the best-known Leap radicals (the so-called four radical kings: Li Jingquan (ranked 89 in 1958) of Sichuan Province, Ke Qingshi (35, an exception to being a lowly ranked FM) of Shanghai, Wu Zhipu (90) of Henan Province and Zeng Xisheng (94) of Anhui Province—were all FMs.Footnote ³⁸ Moreover, while these “radical kings” were indeed ranked just above the cutoff point in 1958, they were anything but passive watchers of the Leap. Yet it takes more than half a dozen zealous provincial leaders to fire all the cylinders of the Leap. Realizing that many of these leaders, if left to themselves, would likely have rested on their laurels, Mao exercised his exceptional authority and bent the rules of the promotion system in order to more effectively influence and manage the political incentives of China's provincial leaders.Footnote ³⁹

Take Li Jingquan of Sichuan Province, for instance. Although Li managed to climb up the rankings from 91st in 1956 to 89th two years later (by dint of the autopilot mechanism), he was nowhere near the rankings normally needed to be considered for promotion to the Politburo. Li was promoted in May 1958 in part to reward his past efforts and loyalty, but perhaps more importantly because Mao, who had the power to exceptionally promote zealous officials regardless of their ranking, wanted to motivate other lowly ranked FMs to pull their acts together. More importantly, Li's promotion was not done in isolation, but arranged jointly with two other radicals—Ke Qingshi (ranked 35th) and Tan Zhenlin (ranked 69th)—which would have sent a loud and clear message to the less-motivated FMs that loyalty to Mao would be duly rewarded. The importance of promoting Li was also noted outside of Sichuan Province, because, in addition to being the first party secretary of Sichuan, he was also in charge of the Southwest bureau apparatus (Xinan Ju), which included the provinces of Yunnan and Guizhou. Li's extraordinarily radical behavior in spearheading the Leap had likely put enormous pressure on the leaders of the other two provinces to keep up.Footnote ⁴⁰

Wu Zhipu of Henan Province is another case in point. Although Wu, who ranked 90th in 1958 (just behind Li), was not as lucky as his Sichuan counterpart, he was still “promoted” from second (i.e., governor) to first party secretary, similarly shortly before the launching of the Great Leap. His otherwise trivial political ascendancy takes on unusual importance in the Leap's context, because Mao had designated Henan Province the “national model” province against which all other Chinese provinces should compare themselves.Footnote ⁴¹