Data Lost and Found

Among the 137 politicians included in its robustness checks, BITTW only obtained full information on all the variables for 40 politicians, on whom it conducted the main analysis. To assuage worries over the meaningfulness of statistical findings with such a small sample, BITTW meticulously applied all the standard confidence-enhancing procedures in quantitative social science, for which it deserved recognition. Nonetheless, being measures of last resort, those procedures are themselves no substitute for a higher-quality sample.

In fact, there is room to improve the sample. Among those excluded from BITTW’s main analysis, there are 28 whose kinship information was procured but whose partisanship information was left blank (Wang Reference Wang2022, Dataset). By taking a closer look at historical records, however, we can retrieve information on the partisan stances of at least 21 politicians among those 28, increasing the sample size from 40 to 61 for the main analysis.Footnote ¹ In particular, most of the information exists in the two major documents BITTW relied on, namely, Xu Zizhitongjian Changbian (The Extended Continuation to Comprehensive Mirror in Aid of Governance, compiled in the twelfth century) (Li Reference Li1995) and Song Shi (The History of Song, compiled in the fourteenth century) (Toghtō 1975). Supplementary as well as corroborative information can be retrieved from other historical materials, such as Sushui Jiwen (Records of Rumors by the River of Su, personal notes Wang Anshi’s nemesis Sima Guang left behind) (Sima Reference Sima1936).

Table 1 presents the expanded list of 26 pro-Reform and 35 anti-Reform politicians for the main analysis, in descending order of LCK. It also groups them in terms of their rounded LCK values, so as to give readers a more intuitive sense of the sample’s configuration. The politicians on whom BITTW did not procure partisanship information are marked with asterisks, and their relevant historical records are provided in the respective footnotes. For the sake of simplicity, hereafter I call BITTW’s original sample for the main analysis the O-Sample, and my expanded sample the E-Sample. Meanwhile, for reasons to be explained later, I also include every politician’s hometown region information in Table 1, coded as “N” if one came from north of “the Huai River, the Tongbai Mountain, the midstream Han River and the Daba Maountains” (Yang Reference Yang1982, 165):Footnote ²³ as “S” if one came from south of that line; and as “W” if one came from the then culturally and economically distinct Southwest, roughly equivalent to present-day Sichuan and Chongqing (Liu Reference Liu1959, 28–9).

Table 1.

Expanded List of Politicians for the Main Analysis, with Information on Hometown Region, in Descending Order of LCK

Table 1. Long description

The table consists of three columns: L C K, Anti-Reform politicians (L C K; N/S/W), and Pro-Reform politicians (L C K; N/S/W). Data is organized in descending order of L C K values.

* L C K approximately 38: Lü Gongzhu (38.33369; N) in the Anti-Reform column.

* L C K approximately 29: Lü Gongbi (29.3579; N) in the Anti-Reform column.

* L C K approximately 15: Lü Hui (15.48733; N) in the Anti-Reform column.

* L C K approximately 13: Zhang Fangping (12.59926; N) in the Anti-Reform column.

* L C K approximately 12: Ouyang Xiu (11.61272; S) in the Anti-Reform column.

* L C K approximately 10: Li Zhongshi (10.04658; N) in the Pro-Reform column.

* L C K approximately 9: Shen Heng (9.122305; S) in the Pro-Reform column.

* L C K approximately 7: Cai Ting (6.789383; N) in the Pro-Reform column.

* L C K approximately 6: Zhang Dun (6.2142; S) in the Pro-Reform column.

* L C K approximately 5: Wang Cun (5.207369; S) in Anti-Reform and Shao Kang (4.904844; S) in Pro-Reform.

* L C K approximately 4: Han Wei (4.235837; N) and Liu Shu (4.007669; S) in the Anti-Reform column.

* L C K approximately 3: Wang Kechen (2.570753; N) in the Pro-Reform column.

* L C K approximately 2: Anti-Reform includes Fan Zhen (2.450677; W) and Han Qi (2.273623; N). Pro-Reform includes Cheng Shimeng (2.409637; S), Zeng Xiaokuan (2.098059; S), and Fan Tang (2.02815; S).

* L C K approximately 1: Anti-Reform includes Hu Zongyu (1.246685; S), Wu Chong (1.069304; S), Chen Jian (0.982577; N), Sun Gu (0.859461; N), Song Minqiu (0.778127; N), Zhao Bian (0.644609; S), and Sima Guang (0.606516; N). Pro-Reform includes Huang Lü (1.082977; S).

* L C K approximately 0: This section contains the largest group. Pro-Reform includes Wang Anshi (0.467142; S), Zeng Bu (0.348693; S), Xu Jiang (0.30112; S), Su Song (0.252845; S), Han Zhen (0.163737; N), Wang Anli (0.136038; S), Wang Gui (0.111833; W), Han Jiang (0.10607; N), Huang Lian (0.084298; S), Yuan Jiang (0.082333; S), Cai Que (0.05989; S), Shen Jichang (0.055021; S), Du Hong (0.042524; N), Deng Runfu (0.040621; S), Cai Yanqing (0.026848; N), and Jian Zhoufu (0.000923; W). Anti-Reform includes Fu Bi (0.423349; N), Tang Jie (0.392212; S), Li Suzhi (0.377345; N), Feng Jing (0.335994; S), Li Chang (0.328793; S), Wen Yanbo (0.305662; N), Chen Xiang (0.26106; S), Lin Dan (0.19842; S), Han Zhongyan (0.186816; N), Li Shen (0.116712; S), Yang Hui (0.108251; W), Teng Yuanfa (0.054217; S), Wu Kui (0.039653; N), Fan Chuncui (0.032859; S), Gong Dingchen (0.028305; N), Wang Zishao (0.009945; N), Qian Xie (0.00897; S), and Chen Yi (0.002011; N).

Note: Politicians with * were omitted in BITTW’s main analysis; LCK: local concentration of kin (Wang Reference Wang2022, Dataset); N: North; S: South; W: Southwest

Notes on Coding

For its main analysis, BITTW adopted a “continuous” approach to coding a politician’s Support for Reform (hereafter SFR): namely, coding consistent support (i.e., across policies and over time) as 1; coding consistent opposition as 0; and coding a mixed attitude by way of the score “averaged across all policies” on which a politician’s view was specified in historical records (Wang Reference Wang2022, 904). As it turned out, BITTW only identified four mixed-attitude politicians in the O-Sample and one outside of it,Footnote ²⁴ whereas all the other politicians were coded as either consistently anti-Reform or consistently pro-Reform. It is likely that BITTW considerably underestimated the prevalence of mixed attitudes; however, as there are quite a few other politicians well known for having either cross-policy mixed attitudes or temporally evolved attitudes toward Reform, both in the O-Sample (such as Wang Zishao)Footnote ²⁵ as well as the rest of the E-Sample (such as Fan Chuncui and Wang Cun),Footnote ²⁶ and among those whose kinship information BITTW did not procure (such as Su Shi and Zheng Xia).Footnote ²⁷ In theory, a more thorough—and much more time-consuming—scraping of additional, rare historical sources (than what I have done to simply expand the sample for the main analysis) might provide us with a longer list of mixed-attitude politicians, each assigned a certain “averaged” SFR score greater than 0 and smaller than 1. But there is no point in doing so, for BITTW’s coding strategy of averaging across specified policies is fundamentally flawed and potentially misleading anyway.

To see why, take Li Chang for example. BITTW assigned him an SFR score of 0.166667, meaning that Li specifically opposed five Reform policies and supported one Reform policy according to available historical records. In reality, however, things were much more complicated. Li used to be Wang Anshi’s close friend and a strong supporter of Reform, and was hand-picked by Wang at the outset of Reform to be a deputy director of the newly established Steering Agency on Finances and Regulations (zhizhi sansi tiaoli si) in charge of devising and financing Reform policies. There, Li offered two pieces of advice to Wang: to slow down the pace of Reform, so as not to make too many enemies at once; and to drop certain ill-thought-out Reform policies from the package, such as the balanced delivery reform (junshufa) and the rural credit reform (qingmiaofa). Both pieces of advice infuriated Wang, however, who was infamously unable to take criticism. Shortly afterwards, Li was purged by Wang from the central government and appointed to various local and ceremonial positions over the next two decades. By the time Emperor Shenzong died, Li had become known as a vocal critic of Reform and was soon recruited by the anti-Reform party, which was incumbent through the Yuanyou Era (1086–1094 CE). Yet, he surprised and alienated his anti-Reform colleagues when they planned to abolish the “hired” labor service reform (muyifa) and restore the pre-Reform system of “assigned” labor service (chaiyifa). According to Li, the older system had been deserted for so long that there was no longer any reliable census data in place to support its implementation; instead, a mixed system should be strived for that would combine the strengths of muyifa and chaiyifa. Footnote ²⁸

Li Chang’s case highlights two problems with BITTW’s coding strategy. First, its averaged SFR scores cannot capture a politician’s attitude change over time, be it toward Reform as a whole or toward any specific Reform policy. For example, we cannot tell from Li Chang’s SFR score (0.166667) that his overall attitude toward Reform changed from support to opposition. Similarly, we cannot tell from Wang Zishao’s SFR score (0) that his view on the rural credit reform changed from support to opposition.

Second, a politician’s SFR score averaged across the policies on which his view was specified in historical records could sometimes misrepresent his overall stance on Reform. After all, Reform was a package of dozens of policies, and a politician might, for a variety of reasons, only engage (or leave archival traces of engagement) in debates over a handful of those policies, even if he also had strong views on other policies or on Reform as a whole. For example, in accordance with BITTW’s coding strategy, Li Chang at the outset of Reform (i.e., excluding his later defense of the labor service reform during the Yuanyou Era, the only available historical record of him defending any specific Reform policy) would have gotten an SFR score of 0 (i.e., consistently anti-Reform), even though, as mentioned above, his criticisms of specific policies at the outset of Reform were made against the backdrop of his overall pro-Reform stance at the time. Similarly, BITTW gave Teng Yuanfa an SFR score of 0.5, leaving the impression that he was a moderate, even though he was actually a staunch opponent of Reform throughout.Footnote ²⁹

I will return to these problems later, when discussing rival hypotheses to BITTW’s theory. For now, my point is that assigning a “continuous” SFR score to a mixed-attitude politician by way of averaging his expressed views on individual Reform policies, as BITTW did, is no better than simply assigning a “dichotomous” SFR score to him based on historical records about his overall attitude toward Reform (i.e., either anti-Reform or pro-Reform), at least when studying the effect of LCK on one’s attitude toward Reform. In other words, for the present purpose of replicating BITTW’s results with the E-Sample, it would be a waste of time either to code the additional 21 politicians “continuously” or to identify additional mixed-attitude politicians in the O-Sample for “continuous” coding. Rather, we should just assign dichotomous SFR scores (0 or 1) to all 61 politicians, including the four politicians coded “continuously” by BITTW.

Status Competition

According to the status-competition explanation, whereas the North had been China’s economic, cultural, and political center for much of Chinese history, by the mid-11th century, it was on the brink of being surpassed, for good, by the burgeoning South. Illustratively, the proportion of chancellors who were Southerners rose from 0% in the early years of the Northern Song dynasty (960–997 CE), to 22.2% during Emperor Zhenzong’s reign (997–1022 CE), to 42.1% during Emperor Renzong’s (1022–1063 CE), and then suddenly to 83.3% under the auspices of Emperor Shenzong (1067–1085 CE), reaching the zenith of Southern political power (followed by a purge of Southern chancellors—and Reform’s repeal—soon after Emperor Shenzong’s death) (Yang Reference Yang1982, 166). The rapidly shifting balance in the North/South status competition created an intense animosity between Northern “old guards” and Southern “new arrivals,” fueling—and being fueled by—the latter’s eagerness to exercise their freshly acquired power and the former’s perception of it as usurpation (Liu Reference Liu1959, 37–8). Now, suppose the nouveau riche are more interested in “marrying up” and building social capital than are the old guard in making peace with its challengers, for whom it cannot help feeling contempt. If so, the status-competition explanation implies the following rival hypothesis to BITTW’s on kinship network dispersal:

Accordingly, insofar as there is an LCK-SFR correlation, it is not because geographically dispersed kinship networks incentivized politicians to support state-building, but because the South/North divide caused the differences in both the geography of kinship networks and attitudes toward Reform. Interestingly, as Table 8 shows, the correlation between the South/North divide and kinship network dispersal is significant in the O-Sample but insignificant in the E-Sample. Suppose, as speculated above, that the LCK-SFR correlation reemerges in a further expanded sample in the future, the same tests on the effects of the South/North divide both on LCK itself (Table 8) and on the LCK-SFR correlation (Table 7) need to be done too, to determine if BITTW’s theory survives the challenge of this rival hypothesis at all.

Table 8.

OLS Estimates of the Correlation between Hometown Region and Geography of Kinship Network (O-Sample vs. E-Sample)

Table 8. Long description

The table compares two groups, O-Sample and E-Sample, regarding the local concentration of kin. The independent variable is Hometown region with North N as the baseline.

* South S: O-Sample shows a coefficient of minus 0.761 with a standard error of 0.369, significant at the p < 0.05 level. E-Sample shows a coefficient of minus 0.498 with a standard error of 0.301.

* Southwest W: O-Sample shows a coefficient of minus 0.840 with a standard error of 0.370, significant at the p < 0.05 level. E-Sample shows a coefficient of minus 0.655 with a standard error of 0.302, significant at the p < 0.05 level.

* Constant: O-Sample is 0.463 with a standard error of 0.361. E-Sample is 0.304 with a standard error of 0.291.

* Observations: O-Sample has 40; E-Sample has 61.

* R-squared: O-Sample is 0.147; E-Sample is 0.067.

All variables are standardized and robust standard errors are provided in parentheses.

Note: The dependent variable is an index of the local concentration of kin. The variable of interest is the politician’s hometown region—North (N), South (S), or Southwest (W)—with North (N) being the baseline. All variables are standardized. Robust standard errors are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.10.

Policy Adaptiveness

According to the policy-adaptiveness explanation, the North and the South were so different in their natural and socioeconomic conditions that no one-size-fits-all Reform package could possibly work in both regions. Meanwhile, the chief architect of Reform, Wang Anshi, who not only grew up in the South but whose magistracy experiences were also confined to South China, was much more versed in Southern conditions than in Northern ones. What’s more, he was by all accounts an adamant and overconfident person unable to take criticism. Consequently, many of his policies were particularly maladapted to the North, provoking vehement Northern opposition. For example, the “hired” labor service reform, with “commutation to cash tax payment” at its core, “worked better in the southern areas where money economy was better developed” but “caused considerable hardship in the northern areas” (Liu Reference Liu1959, 109). As the Chinese proverb goes: tangerines grown south of the Huai River are sweet tangerines, while those grown in the north, due to their different shuitu (“water-and-earth”), become bitter trifoliate oranges.

Compared with other explanations, the policy-adaptiveness explanation is particularly attuned to the nuanced attitudes many politicians held toward Reform: a politician would support one Reform policy and oppose another if the former was adaptive, and the latter was maladaptive, to the local conditions with which he was versed in; and he might change his view on a particular Reform policy, or on Reform as a whole, when he acquired knowledge on the local conditions with which he had not been versed. Meanwhile, it also makes better sense of a fact BITTW did not mention (and could not account for), that many leading opponents of Reform (such as Fu Bi, Han Qi, and Ouyang Xiu) had been whole-hearted supporters of the thwarted Qingli New Deal (1043–1045 CE) two decades before, seeking also to strengthen state capacity. BITTW’s theory implies—quite implausibly—that the “Qingli generation” politicians turned against state-building two decades later because, as time went by, their kinship networks became more concentrated geographically. By contrast, according to the policy-adaptiveness explanation, they opposed Reform not because they opposed state-building per se, but because they deemed it a misinformed package of state-building policies, which its bullheaded architect pushed hastily through. In a nutshell, “anti-Reform” should not be equated with “anti-reform” (i.e., opposing reforms that strengthen state capacity).

Like the status-competition explanation, the policy-adaptiveness explanation begets a rival hypothesis to BITTW’s on kinship network dispersal:

A geographically dispersed kinship network could increase a Northern politician’s exposure to Southern local knowledge, as well as a Southern politician’s exposure to Northern local knowledge. Given that Reform was more adaptive to Southern conditions than to Northern conditions, a Northerner with a dispersed kinship network was likelier than a Northerner with a concentrated kinship network to acknowledge that at least some Reform policies worked in certain parts of China, and a Southerner with a dispersed kinship network was likelier than a Southerner with a concentrated kinship network to acknowledge that at least some Reform policies did not work in certain parts of China.Footnote ³³ Hence the hypothesis.

It would be a daunting task to test this hypothesis quantitatively, however. To begin with, we ideally must know—and control for—when and where a politician served as a local magistrate, which, like having a dispersed kinship network, might expose him to local conditions different from where he grew up and affect his view on Reform. Sans the influence of one’s kinship network, the timing of his magistracy experience affects how well-versed he already was in various local conditions before expressing or changing his view on Reform (or on any specific Reform policy), and the whereabouts of those experiences could be relevant to his policy-specific attitudes.Footnote ³⁴ Relatedly, we also need as detailed information as possible on politicians’ policy-specific attitudes, as well as on their attitude changes over time. Yet, historical records of the Song dynasty often specified only where, but not when, a politician worked as a magistrate; and as mentioned earlier, it is likely that the data gathered on policy-specific attitudes and attitude changes would also be highly incomplete, even if we were to increase exponentially the time and effort spent on scraping additional information from rare historical sources. In other words, the missing-data problem looms even larger for the quantification of Rival Hypothesis 2 than for BITTW.