How does regional GDP manipulation affect livelihood investment?

The paper examines how manipulation of regional GDP statistics by local governments affects residents' welfare through changes in livelihood investment. In contexts where officials are evaluated and promoted based on economic performance, political incentives can lead to distorted economic indicators, including GDP, crowding out spending on public goods such as healthcare, education, and social security. China provides a salient setting given long-standing debates on the reliability of GDP data, lagging social service development relative to economic growth, and a promotion system emphasizing GDP performance. Using satellite night-time lights to approximate true economic activity, the study asks whether higher levels of GDP manipulation are associated with lower shares of fiscal expenditure devoted to livelihood, and whether this relationship varies with officials’ tenure, hometown status, and macroeconomic conditions. The study aims to contribute to understanding the political economy trade-off between growth signaling and welfare investment.

Prior work highlights the centrality of GDP growth in Chinese officials’ promotion prospects, raising concerns over data reliability and manipulation. Studies document discrepancies in China’s GDP data and strategic behavior by local governments to meet growth targets. The literature also notes trade-offs between growth-focused projects and provision of public goods with delayed payoffs. Political budget cycles and motivations to maintain social stability or win support can further shape fiscal decisions. Geographic identity and regional favoritism can influence resource allocation toward officials’ hometowns. Based on this literature, the authors develop three hypotheses: Hypothesis 1: Greater regional GDP manipulation is associated with lower livelihood investment. Hypothesis 2: The negative impact of GDP manipulation on livelihood investment is more pronounced in the early years of local officials’ tenures (political cycle effect). Hypothesis 3: The adverse effect of GDP manipulation on livelihood investment is weaker when officials serve in their hometowns (regional favoritism attenuates the effect).

Data and sample: The study covers 30 mainland Chinese provinces and municipalities (excluding Hong Kong, Macao, Taiwan, and Tibet) over 1998–2013. Livelihood investment is defined as the share of education, social/employment security, and healthcare expenditures in general public budget expenditure (SOCISRV). Fiscal classification adjustments around 2007 are harmonized. Economic data and officials’ characteristics are from the CSMAR database, China Statistical Yearbook, and provincial yearbooks. Night-time light data: DMSP/OLS night lights (1992–2013) from NOAA are calibrated to address (i) saturation (upper limit DN=63) using radiance-calibrated correction data and relationships between stable lights and radiance data; (ii) inter-annual discontinuity via pseudo-invariant feature (PIF) calibration (power-function intercalibration using Mauritius, Puerto Rico, Okinawa); (iii) within-year cross-satellite inconsistencies via averaging where applicable; and (iv) sensor aging/assumed consistent growth issues via corrections for anomalous declines following Elvidge et al. (2009). Measurement of GDP manipulation: Following Henderson et al. (2012), official GDP growth is regressed on growth in calibrated light brightness per unit area with province and year effects to obtain a fitted growth from lights. A composite estimate of true GDP growth ĝ is constructed by weighting official GDP growth and light-imputed growth with ρ=0.586 (per Henderson et al., Xu et al., 2015). GDP manipulation GDPDIS is defined as official GDP growth minus estimated true GDP growth; positive values indicate official growth exceeds true growth. Baseline empirical model: SOCISRV_it = α0 + α1 GDPDIS_{i,t−1} + β Controls_it + φ_i + γ_t + η_it. Controls include: development level (DEV, log real GDP per capita), industrial structure (INDUS, share of secondary and tertiary industries), fiscal self-sufficiency (SLFSUFFI), fiscal decentralization (DECENT), demographic demand (DEMAND, dependency ratio), FDI, and officials’ characteristics (governor and party secretary age, gender, education, local/non-local origin). Fixed effects for province and year are included in the preferred specification. Heterogeneity analyses: Subsamples by governor tenure (TENURE) to test political cycle effects; interactions with officials’ hometown indicators (LOCAL_sz, LOCAL_sj) to test regional favoritism. Robustness: Dynamic panel estimations (DIF-GMM and SYS-GMM) adding lagged SOCISRV; alternative dependent variables (log total social service expenditure deflated by CPI, and log per-capita social expenditure); exclusion of resource-rich provinces; and splits by pre- and post-global financial crisis periods (1998–2007 vs. 2008–2013).

• Descriptive evidence: SOCISRV averages 31.69% (SD 5.03, range 19.5%–43.23%). SOCISRV and GDPDIS are negatively correlated (r = -0.125, significant). Nationally (1993–2012), official GDP growth averages 10.20% versus estimated true growth 10.13%, implying an average manipulation rate around 0.54%. Some regions (e.g., Inner Mongolia, Tianjin) show larger positive GDPDIS; in many central/western regions, true growth exceeds official figures (negative GDPDIS). - Baseline regressions (Table 5): GDPDIS is significantly negatively associated with SOCISRV. OLS coefficient: -1.103 (0.262); Fixed-effects coefficient: -0.483 (0.191). Control variables behave plausibly; province and year fixed effects models have high within R^2 (0.844). - Political cycle heterogeneity (Table 6): The negative effect is concentrated in earlier tenure. For governors with TENURE ≤ 3 years, GDPDIS coefficient = -0.699 (0.248); for TENURE > 3 years, coefficient = -0.312 (0.347) (not significant). - Regional favoritism (Table 7): Interaction terms indicate attenuation when officials serve in their hometowns. Governor hometown interaction GDPDISLOCAL_sz: -0.726 (t≈-2.35, p<0.05), implying stronger negative effect when governors are non-local; Party secretary hometown interaction GDPDISLOCAL_sj: -1.676 (t≈-2.42, p<0.05), indicating the negative effect is significantly smaller when party secretaries are local. - Dynamic panels (Table 8): Lagged SOCISRV is positive and significant (0.384–0.434). GDPDIS remains negative and significant: DID-GMM -0.713 (0.213)*; SYS-GMM -0.659 (0.229)**. AR(1) present, AR(2) absent; Sargan tests acceptable. - Alternative outcome measures (Table 9): Results hold using log total social service expenditure (SOCISRV_1) and log per-capita social expenditure (SOCISRV_2); GDPDIS coefficients are negative and significant at conventional levels. - Excluding oil-producing provinces (Table 10): Negative effect persists across specifications. - Macroeconomic conditions (Table 11): The study reports that during economic slowdown (2008–2013), the contraction in livelihood investment associated with GDP manipulation is more pronounced, consistent with heightened fiscal stress; split-sample estimates are directionally more negative in the later period. Overall, results support Hypotheses 1–3: higher GDP manipulation is linked to lower livelihood investment, especially early in officials’ tenures, and the effect is attenuated when officials serve in their hometowns; it is also stronger in downturns.

The findings indicate that when local governments inflate GDP figures to signal strong performance, fiscal resources are reallocated toward short-term, growth-visible projects at the expense of public goods with longer-term payoffs, such as education, healthcare, and social security. This diversion likely reflects both political incentives (promotion systems emphasizing GDP) and fiscal dynamics (over-optimistic revenue projections and imbalances following manipulation). The stronger negative association early in officials’ tenures aligns with political budget cycle logic: officials prioritize quick, demonstrable growth to build promotion records, reducing social service shares. The attenuation when officials serve in their hometowns suggests regional favoritism can redirect resources toward local welfare, partially offsetting the manipulation-welfare trade-off. During macroeconomic slowdowns, tighter fiscal constraints magnify the crowding-out of livelihood spending, intensifying welfare costs of manipulation. Collectively, the evidence links data credibility to welfare outcomes, highlighting how manipulation-induced distortions propagate into the composition of public expenditure and ultimately residents’ welfare.

The study shows that regional GDP manipulation significantly reduces the share of fiscal resources allocated to livelihood investments in China. The negative effect is most pronounced in the early years of officials’ tenures and is mitigated when officials serve in their hometowns. Evidence also suggests that in periods of economic downturn, manipulation-related contraction in social service spending is exacerbated. These results extend research on both GDP manipulation and public expenditure composition by connecting data credibility to welfare-oriented fiscal outcomes. Policy implications include strengthening statistical systems (e.g., quasi-vertical management of provincial GDP statistics under the National Bureau of Statistics), leveraging technological tools (such as satellite-based validation) to reduce information asymmetry, and reweighting cadre evaluations to emphasize social services and livelihood outcomes alongside growth metrics. Doing so could help curb manipulation incentives and rebalance fiscal priorities toward residents’ welfare. Future work could explore micro-level welfare impacts, heterogeneity across policy domains within social services, and the effects of post-2014 institutional reforms.

• Measurement: Night-time lights, while informative, are an imperfect proxy. Despite calibration (saturation correction, inter-annual alignment, sensor aging adjustments), residual measurement error may remain. Differences between light-imputed and official GDP are not solely due to manipulation (e.g., sectoral composition, energy intensity, geography). - External validity over time: The sample ends in 2013 to avoid confounding from major 2014 reforms; results may differ under subsequent institutional changes. - Omitted variables and identification: Although extensive controls, fixed effects, robustness checks, and GMM are used, unobserved factors could still influence both manipulation and spending composition. - Data constraints: Some provincial fiscal and official characteristic data required supplementation; accounting changes around 2007 necessitated harmonization, potentially introducing noise. - Macroeconomic split precision: The downturn-period estimates are directionally consistent but may have limited precision in split samples.