Decomposing Firm Performance

There is an ongoing debate in the literature concerning the extent to which business, corporate, and industry factors explain variance in firm performance (Karniouchina, Carson, Short, & Ketchen, Reference Karniouchina, Carson, Short and Ketchen2013; McGahan & Porter, Reference McGahan and Porter1997; Roquebert, Phillips, & Westfall, Reference Roquebert, Phillips and Westfall1996; Rumelt, Reference Rumelt1991). Industrial organization (I.O.) theorists argue that a firm's performance is determined mainly by the industry in which it operates (Hall & Weiss, Reference Hall and Weiss1967; Porter, Reference Porter1980; Scherer & Ross, Reference Scherer and Ross1990), with the industry being the primary source of performance variance. Alternatively, the RBV suggests that diversity in firms' resources and capabilities determines firm performance (Barney, Reference Barney1991; Montgomery & Wernerfelt, Reference Montgomery and Wernerfelt1988; Wernerfelt, Reference Wernerfelt1984), which supports the business unit as the primary source of performance variance.

Several studies have endeavored to delineate the relative importance of business, corporate, and industry factors in determining firm performance (see Vanneste (Reference Vanneste2017) for a detailed review). Previous studies show that industry effect ranges from 1 to 20% of the variance in firm performance, while business unit effect ranges from 20 to 40%. For example, McGahan and Porter (Reference McGahan and Porter1997) found that business and industry factors accounted for 32 and 19% of the total variance, respectively. Other studies replicated similar results, leading to a consensus that business is a more significant factor than the industry in explaining performance variance. In a meta-analysis using 18 samples from 16 studies, Vanneste (Reference Vanneste2017) found that industry accounts for 8%, corporate accounts for 14%, and business accounts for 36% of the total variance.

Besides industry, corporate, and business segment effects, scholars also examined other factors that explain firm performance, including ownership form (Fitza & Tihanyi, Reference Fitza and Tihanyi2017; Xia & Walker, Reference Xia and Walker2015), CEO or board membership (Krause, Li, Ma, & Bruton, Reference Krause, Li, Ma and Bruton2019; Quigley & Hambrick, Reference Quigley and Hambrick2015), business group (Chang & Hong, Reference Chang and Hong2002; Sharapov, Kattuman, Rodriguez, & Velazquez, Reference Sharapov, Kattuman, Rodriguez and Velazquez2021), country (Hermelo & Vassolo, Reference Hermelo and Vassolo2012; McGahan & Victer, Reference McGahan and Victer2010; Tong, Alessandri, Reuer, & Chintakananda, Reference Tong, Alessandri, Reuer and Chintakananda2008), and alliance network (Kumar, Liu, & Zaheer, Reference Kumar, Liu and Zaheer2022). Previous studies focused mainly on US firms and utilized the Compustat data, while some have considered other countries, including Korea (Chang & Hong, Reference Chang and Hong2002), Japan (Makino et al., Reference Makino, Isobe and Chan2004), and Sweden (Short, McKelvie, Ketchen, & Chandler, Reference Short, McKelvie, Ketchen and Chandler2009). The discrepancy in results is minimal across different country samples. Few studies have focused on emerging markets to explore the determinants of performance variance. One exception is Xia and Walker (Reference Xia and Walker2015), which found that ownership type significantly impacts firm performance and interacts with geography and time in a sample of Chinese manufacturing firms from 1998 to 2007.

In previous studies, firm profits, namely return on assets (ROA), has been the primary performance measure. However, ROA faces its limitations (Hawawini, Subramanian, & Verdin, Reference Hawawini, Subramanian and Verdin2003). For example, it does not account for the cost of capital and fails to adequately explain the value created in a firm. Some studies applied alternative performance measures, including Tobin's Q (Wernerfelt & Montgomery, Reference Wernerfelt and Montgomery1988), market share (Chang & Singh, Reference Chang and Singh2000), return on sales (Makino et al., Reference Makino, Isobe and Chan2004), growth option value (Tong et al., Reference Tong, Alessandri, Reuer and Chintakananda2008), economic profit, and book-to-market value (Hawawini et al., Reference Hawawini, Subramanian and Verdin2003), to overcome the limitations of ROA. However, given similar results across different performance measures, the field has adopted ROA as a gold-standard performance measure in decomposing performance variance. Studies have hardly considered growth as a performance measure, except for Short et al. (Reference Short, McKelvie, Ketchen and Chandler2009), who used growth to examine the performance of new ventures in Sweden. Using hierarchical linear modeling, they found that firm factors are much more important than industry factors in explaining growth variance in new ventures and established firms.

There has also been a debate on how to conduct the analysis or what method to use to decompose performance variance. Studies have adopted different parametric approaches, starting with the analysis of variance (ANOVA) (Rumelt, Reference Rumelt1991; Schmalensee, Reference Schmalensee1985) and component-of-variance (COV) analysis (Makino, Lau, & Yeh, Reference Makino, Lau and Yeh2002; McGahan & Porter, Reference McGahan and Porter1997). Although these methods exhibit desirable strengths, such as showing the relative importance of different factors, they suffer from critical limitations; for example, they are limited in handling random year and interaction effects (Guo, Reference Guo2017). Later studies (Guo, Reference Guo2017; Hough, Reference Hough2006; Misangyi et al., Reference Misangyi, Elms, Greckhamer and Lepine2006) have applied multilevel modeling (MLM) to overcome these limitations. MLM can better handle cross-classified factors and interactions, estimate unbalanced data efficiently, and allow for categorical and continuous variables.

In sum, previous studies generally present consistent findings on different determinants of firm performance, regardless of research method, sample, country, and modeling. While the findings remain consistent across different performance measures, there is still a lack of understanding of the determinants of growth variance across firms. Moreover, revisiting these issues in emerging market contexts is necessary while incorporating their unique characteristics. This would lead to new insights into firm performance and improve the overall understanding of the determinants of performance variance.

Firm Resource and Performance

The RBV is a dominant theory explaining firm profits and growth (Barney, Reference Barney1991; Penrose, Reference Penrose1959). Although both Penrose (Reference Penrose1959) and Barney (Reference Barney1991) highlighted the role of firm resources in determining growth, they focus on different types of resources. Penrose (Reference Penrose1959) argued that excess and versatile resources lead to firm growth. Versatile resources refer to resources that offer a broad range of potential services (Penrose, Reference Penrose1959). As Penrose (Reference Penrose1955: 539) posits, ‘It becomes clear that the flexibility and versatility of its resources are the important factors governing the possibilities of its expansion. As long as profitable production opportunities are available anywhere in the economy, a firm can take advantage of them if its resources are versatile’. The redeployment of versatile resources enables firms to pursue new applications for the resources, thus pushing for growth.

Barney (Reference Barney1991) argued that VRIN resources lead to sustained competitive advantage and, thus, long-term above-average returns. VRIN resource is directly related to firm competitiveness and thus profitability. VRIN resources could also explain firm growth because firms could exploit VRIN resources in related areas and thus achieve growth consistent with the logic of related diversification (Markides & Williamson, Reference Markides and Williamson1994).

Comparing VRIN resources and versatile resources, we can see that there are overlaps between the two. There are two types of versatile resources: tradable between firms and deployable within a firm (Nason & Wiklund, Reference Nason and Wiklund2018). The former includes cash, commodities, and generic human resources (Kim & Bettis, Reference Kim and Bettis2014; Mishina, Pollock, & Porac, Reference Mishina, Pollock and Porac2004). The latter includes uniquely developed resources with a broader range of use, such as brands and technologies (Anand & Delios, Reference Anand and Delios2002; Danneels, Reference Danneels2002). While the latter type of versatile resources is similar to VRIN resources, the former type is not. Meanwhile, not all VRIN resources can be applied to a broad range of areas, such as specific technical knowledge of employees. The combination of versatile and VRIN resources could help a firm achieve sustained high performance.

This study focuses on the difference between VRIN and versatile resources. In particular, their impact on firm performance is different. Versatile resources can explain a broader range of growth than VRIN resources. Versatile resources that are easily tradable among firms, such as cash, could drive firm growth by offering a wide range of potential use in related and unrelated areas. Moreover, since versatile resources have lower transaction costs than VRIN resources, they enable firms to quickly shift their growth strategy to pursue opportunities in external environments (Nason & Wiklund, Reference Nason and Wiklund2018). On the contrary, VRIN resources only explain a small range of growth paths a firm could pursue (exploiting VRIN resources in current and related areas) (Barney, Reference Barney1991). As a result, versatile resources are more relevant than VRIN resources in explaining firm growth. In a meta-analysis of 113 studies from 1987 to 2011, Nason and Wiklund (Reference Nason and Wiklund2018) found that versatile resources considerably impact firm growth more than non-versatile resources, but it does not matter for growth whether a resource is VRIN or not.

While versatile resources are more relevant in influencing firm growth, VRIN resources are more relevant in influencing firm profit. The core argument of RBV is that VRIN resources create competitive advantages and, thus, above-average returns (Barney, Reference Barney1991; Wernerfelt, Reference Wernerfelt1984). Versatile resources, especially those tradeable among firms, do not necessarily generate competitive advantages and are thus weakly related to firm profit. Compared to VRIN resources, versatile resources, especially those tradeable among firms, such as cash, are less firm-specific. Therefore, we expect the firm effect to influence firm profit than growth substantially.

Firm Growth in Emerging Markets

Stage 1: Comparing Firm Profit and Growth in the US

This stage examines the determinants of firm profit and growth in one of the largest developed countries, the US. We start by using McGahan and Porter's (Reference McGahan and Porter1997) study design. This study is influential within this stream of literature, with more than 2,800 Google Scholar citations by November 2022. Many subsequent studies were anchored by this study (Chang & Singh, Reference Chang and Singh2000; Fitza & Tihanyi, Reference Fitza and Tihanyi2017; Guo, Reference Guo2017). We use their sample and method to show the difference in firm performance between profit and growth in the US when we adopt the sample and method used by most prior studies on decomposing firm performance.

Sample and Measures

Like McGahan and Porter (Reference McGahan and Porter1997), we collected data from the Compustat Business Segment database. We screened the raw data (367,885 observations) according to several exclusion criteria: financial institutions, firms that were the only ones in their industry in a given year, segments with only one year of observations or with assets and sales less than $10 million, and firms with a missing primary SIC code and missing profit data. We also excluded the first year of observations since we applied lagged performance data. The final US sample included 69,483 segments from 13,992 firms from 1981 to 1994. Table 1 presents the screening process and the summary statistics of the data. Although our sample is slightly larger than that of McGahan and Porter (Reference McGahan and Porter1997), the two samples have similar levels of profit and variance. We measured profit as the ratio of operating income to identifiable assets as a percentage.

Table 1. Screening and summary statistics of US samples

Following McGahan and Porter (Reference McGahan and Porter1997), we first reproduced similar results by estimating equation (1) using the COV estimation method

(1)$$\sigma _R^2 = ( 1 + \rho ^2) \sigma _r^2 + ( 1-\rho ^2) ( {\sigma_\alpha^2 + \sigma_\beta^2 + \sigma_\emptyset^2 } ) + 2( {1-\rho^2} ) C_{\alpha \beta } + \sigma _\omega ^2 , \;\;$$

where $\sigma _R^2$ is the variance of profit, ρ is the rate of persistence, $\sigma _r^2$ is the population variance of year, $\sigma _\alpha ^2$ is the variance of industry, $\sigma _\beta ^2$ is the variance of corporate parents, $\sigma _\emptyset ^2$ is the variance of the segment, 2C _αβ is the population covariance between industry and corporate effects, and $\sigma _\omega ^2$ is the variance of error. Equation (2) generates t rate of persistence ρ as follows:

(2)$$r_{i, k, t} = {\rm \rho }r_{i, k, t-1} + ( {1-\rho } ) \mu + \gamma _t-( {1-\rho } ) \gamma _{t-1} + ( {1-\rho } ) \alpha _i + ( {1-\rho } ) \beta _k + ( {1-\rho } ) \emptyset _{i, k} + \omega _{i, k, t}, \;\;$$

where r _i,k,t is the accounting profit in year t of the corporate parent k's business in industry i, μ is the average profit over the entire period for all business segments, γ _t is the difference between μ and the average profit of all business segments in year t, α _i is the increment to profit associated with participation in industry i, β _k is the increment to profit due to membership in a diversified corporate parent k, $\emptyset _{i, k}$ is the increment to profit associated with the specific situation of the corporate parent k's business segment i given other factors, and ω _i,k,t is the error term.

Results

The results are summarized in Table 2. Model 1 shows McGahan and Porter's (Reference McGahan and Porter1997) original results, broadly consistent with our results of replication (Model 2). In Model 3, we applied MLM estimation instead of COV. We used the lmer function in the lme4 package for R, which provides various functions to fit and analyze linear mixed models, generalized linear mixed models, and nonlinear mixed models (Bates, Machler, Bolker, & Walker, Reference Bates, Machler, Bolker and Walker2014). The model estimation followed the restricted maximum likelihood method. Model 3 explains almost 80% of the total variance, compared to around 50% in Model 1. Accordingly, all factors except the segment explained more variance than before, with the industry effect increasing from 18.68 to 29.07%, corporate effect from 4.33 to 11.44%, and year effect from 2.39 to 9.21%. Segment, however, decreased slightly from 31.71 to 27.46%. The increase in the year effect is consistent with the findings of Guo (Reference Guo2017), which showed that the year effect increases in the MLM estimation.

Table 2. Results using US samples

Next, we added industry interactions with year and corporate parent. The results of Model 4 in Table 2 show an interaction effect only between year and industry, at 11.38%. Compared with the results in Model 1, the significant differences were the increase in year effect (from 2.39 to 12.68%) and the interaction between industry and year. Comparison with the results of Model 3 implies a potential dynamic effect, the changing roles of industry over time.

We then extended McGahan and Porter (Reference McGahan and Porter1997) by changing the dependent variable to sales growth (Models 5–7 of Table 2). We first used COV and the independent variables McGahan and Porter (Reference McGahan and Porter1997) used, with the results summarized in Model 5 of Table 2. Compared with Model 2, the corporate effect was larger (4.90 vs. 9.45%), while the business segment effect (30.89 vs. 28.55%) was smaller.

Next, we used MLM to estimate the same model; the results are summarized by Model 6 of Table 2. Compared with the results of Model 3, the segment effect decreased from 27.46 to 18.68%. Following Quigley and Hambrick (Reference Quigley and Hambrick2015), we conducted Fisher's Z-test to test the variance difference (Fisher, Reference Fisher1915). The tests suggest significant differences in the segment effect between Model 3 and Model 6 (P-value < 0.000). The differences suggest that firm-specific factors are less critical in driving sales growth than profitability for US firms.

Finally, we added the year's interaction terms with industry and corporate parent. The results are shown in Model 7 of Table 2. Similar to the results of Model 4, only the interaction between industry and year was substantive, at 11.36%. Compared with the results of Model 4, we found a consistent result that the segment played a less critical role in driving growth (23.92%) than profit did (28.48%) (p < 0.000).

In the next stage, we also analyze data from 1995 to 2012 for the same period for the US and Chinese samples. The results are summarized in Models 8 and 9 of Table 2. We find a consistent difference in the role of the segment in determining growth (27.70%) and profit (33.67%) (p < 0.000).

To summarize, the analyses using the US sample highlight a few points. First, even for the US sample, the determinants of profit and sales growth differ: segment or firm effect is more critical in driving profit than growth. Therefore, we should not treat sales growth and profit as interchangeable performance measures, requiring a separate investigation for sales growth. These results could be driven by the different types of resources driving profit and growth. Second, the results of our US sample are broadly consistent with previous studies, except for the absence of substantive corporate–industry interaction in previous studies. The year effect was much larger when we adopted the MLM technique. This finding is consistent with Guo (Reference Guo2017), who also found a larger year effect using the MLM technique. Lastly, the results show that the research method matters in analyzing performance variance. The MLM technique can reflect the dynamic effect better while substantially improving the total explained variance (e.g., around 80% for MLM and 50% for COV against profit).

Stage 2: Performance Variance for Chinese Firms

This stage examines the determinants of both firm profit and growth in one of the largest emerging markets: China. We employed the same method as the first stage but with a different sample. We intend to show that the difference between profit and growth determinants persists in emerging markets. We include ownership identity as a determinant because previous studies have shown that it influences firm performance (Fitza & Tihanyi, Reference Fitza and Tihanyi2017; Xia & Walker, Reference Xia and Walker2015). We also include the region as a determinant because previous studies have shown considerable intra-country variance in large emerging markets such as China (Cole, Elliott, & Zhang, Reference Cole, Elliott and Zhang2009; Zhong, Lin, Gao, & Yang, Reference Zhong, Lin, Gao and Yang2019). Such differences in external environments also influence firm performance (Xia & Walker, Reference Xia and Walker2015). The region indicates the province where a firm is located.

Sample and Measures

The Chinese firm data is from the Database of Industrial Firms (DIF) compiled annually by the Chinese National Bureau of Statistics, including 4,743,693 manufacturing firms from 1995 to 2012. Because of the comprehensive coverage of the dataset, previous studies have used it to explore various issues, such as the impacts of market liberalization (Park, Li, & Tse, Reference Park, Li and Tse2006), competition (Chang & Xu, Reference Chang and Xu2008), and also performance variance decomposition (Xia & Walker, Reference Xia and Walker2015).

We screened the raw data following the same procedure we did for the U.S. sample. The final sample included 2,554,996 observations from 523,172 firms from 1995 to 2012. This is much larger than the typical sample in previous studies, which generally included less than 100,000 observations. The sample covered 821 industries over 18 years. The average ROA was 10.17%, and the variance was 575.72%. The profit level was similar to the US sample, but Chinese firms have a much higher variance in profit and show high volatility.

The Chinese dataset provides information on the ownership identity of each firm. The most representative ownership identity was private limited liability firms, which accounted for about one-third of the total firms. State-owned businesses accounted for 6.2% of the total observations. Joint ventures with Hong Kong (H.K.), Macau, Taiwan, and foreign firms accounted for about 10%. Foreign firms, including H.K., Macau, and Taiwan, account for about 12%. The left side of Table 3 summarizes the distribution of ownership identities in the Chinese sample.

Table 3. Ownership identities in Chinese and global samples

Results

We first estimated McGahan and Porter (Reference McGahan and Porter1997)'s model using the Chinese sample and MLM. The results are summarized in Model 1 of Table 4. Because the Chinese data did not include segment-level data, we only had the firm effect, which combines corporate and segment effects.

Table 4. Mixed model results of Chinese firms

Next, we added ownership identity and region as explanatory variables in Model 2 of Table 4. Ownership accounted for 12.19% of the total variance, and the region accounted for 7.87%. Ownership identity and region are important factors in explaining performance variance in China. In Model 3, we added the interaction between year and other independent variables. The interaction between industry and year accounted for 10.25% of the total variance, similar to the US sample results. The interaction between ownership identity and year accounted for 9.11% of the total variance, and the interaction between ownership identity and year accounted for 4.39%.

We then changed the dependent variable to sales growth and found similar results, as summarized in Models 4–6 of Table 4. Again, ownership identity accounted for 14.38% of the total variance, and region accounted for 9.43% of the total variance when entered in Model 5. In Model 6, ownership identity and its interaction with year accounted for 23.71% of the total variance, and region and year interaction accounted for 12.07% of the total variance. The results support the argument that ownership identity and region matter in determining firm growth in emerging markets.

The analyses using the Chinese sample suggest several points. First, the firm effect is more critical in determining profit than growth in emerging markets such as China. Models 3 and 6 in Table 4 show that the firm effect is larger for profit (15.78%) than for sales growth (10.36%) (p < 0.000). This result is consistent with that in Stage 1 using the US sample. Second, the industry effect (21.43%) is more important than the firm effect (10.67%) in driving firm growth in China, suggesting a more significant role of external factors in firm growth in emerging markets. Third, ownership identity and region matter in determining both firm profit and growth. Ownership, region, and interactions with year together accounted for more than 30% of the total variance for profit and sales growth.

Stage 3: Performance Variance for Emerging and Developed Markets

This stage further confirms the previous findings and compares the determinants of firm profit and growth in emerging and developed markets. The results of the two previous stages are incomparable due to different sampling procedures. In this stage, we draw data from the same dataset, thus ensuring the comparability of firms from different countries.

Sample and Measures

We decomposed the profit and sales growth variances in a sample of global firms from multiple countries. The global firm sample was from Osiris, which includes information on publicly listed companies worldwide. The dataset is from Bureau van Dijk (BvD), including about 80,000 publicly listed firms worldwide. This data has been broadly used in the field, with detailed and rich information beyond typical financial reports (Kalasin, Dussauge, & Rivera-Santos, Reference Kalasin, Dussauge and Rivera-Santos2014; Surroca, Tribo, & Zahra, Reference Surroca, Tribo and Zahra2013).

The final sample included observations from 1995 to 2012 consistent with the Chinese sample, covering 331 industries over 18 years. The average ROA was 3.66%, and the variance was 15.30. Figure 1 compares the distribution of sample companies by year across the four samples (McGahan and Porter (Reference McGahan and Porter1997), our US, Chinese, and global samples). Table 5 shows the descriptive statistics for all samples.

Figure 1. Sample distribution by year and country contexts

Table 5. Descriptive statistics for different samples

Notes:

^† The number is the P-value of the skewness test. The null hypothesis is that a variable is normally distributed.

^‡ The number of McGahan and Porter (Reference McGahan and Porter1997) and our US sample is at the segment level; the number of Chinese and global samples is at the firm level.

Osiris provides information on the ultimate owner of each firm, and it traces the shareholder with the highest direct or total percentage of ownership. Among the different types of ultimate owners, corporate accounted for most firms, and families were the second largest group, accounting for less than 10%. The right side of Table 3 summarizes the distribution of different ownership identities among global firms.

Results

Using the global sample, we first estimate models, including country and ownership effects. Models 1 and 2 in Table 6 present the results. The dependent variable is profit in Model 1 and growth in Model 2. Again, we see differences in the determinants of growth and profit: firm effect was more critical for profit (27.31%) than sales growth (20.39%) (p < 0.000). Ownership and its interaction with year explained more than 10% of the total variance in profit and growth, and country and its interaction with year explained around 20% of the total variance in profit and growth.

Table 6. Mixed model results of global firms with ownership type

Next, we split the sample into non-OECD (Organization for Economic Co-operation and Development) and OECD countries. OECD countries are generally regarded as developed countries, while non-OECD countries as emerging markets (Boehmer, Nash, & Netter, Reference Boehmer, Nash and Netter2005; Dewan & Kraemer, Reference Dewan and Kraemer2000; Drori, Yonk Suk, & Meyer, Reference Drori, Yonk Suk and Meyer2006). The results of the subsamples are summarized in Models 3–6 of Table 6. These results largely confirm our previous findings.

First, for both OECD and non-OECD countries, the firm effect is more critical in determining profit than growth. For non-OECD countries, the firm plays a more critical role in determining profit (15.11%) than sales growth (9.21%) (p < 0.000). For OECD countries, the firm effect is also larger for profit (28.99%) than sales growth (22.30%) (p < 0.000).

Second, if we focus on the determinants of firm growth, for non-OECD countries, the industry played a more critical role than the firm in determining sales growth (13.99 vs. 9.21%) (p < 0.000). But it is not the case for OECD countries (10.70 vs. 22.30%).

Third, the firm effect is more critical in determining performance in OECD countries than non-OECD countries. For OECD countries, firm effect explains 28.99% of the total profit variance and 22.30% of the total growth variance. For non-OECD countries, the corresponding numbers are 15.11 and 9.21% (p < 0.000), respectively.

Besides the difference in the role of firm effect between OECD and non-OECD countries, industry and year effects are also different. The year effect is larger for non-OECD countries (12.14% for profit and 12.62% for growth) than OECD countries (8.23% for profit and 9.87% for growth) (p < 0.000), implying higher environmental volatility in emerging markets. The industry effect is larger in non-OECD countries (12.27% for profit and 13.99% for growth) than in OECD countries (10.14% for profit and 10.70% for growth) (p < 0.000), consistent with McGahan and Victer (Reference McGahan and Victer2010), who found that industry effects tend to be higher in developing countries. The country effect is also larger for non-OECD countries than for OECD countries.

Finally, ownership identity and country matter for both non-OECD and OECD countries. Ownership identity and its interaction with year explain more than 10% of the total variance for non-OECD and OECD countries. Country and its interaction with year explain more than 15% of the total variance for non-OECD and OECD countries.

We also perform additional tests by adopting different ways to classify developed countries and emerging markets. First, we employ the United Nations Human Development Index (HDI) to classify developed countries. The HDI assigns an index ranging from 0 to 1 for a country's human development, such as education, health, and life expectancy. The top 25% of HDI consists of developed countries. Second, we used the World Bank's high-income country list to identify developed countries. The results in Models 7–14 in Table 6 confirm the previous findings for different classifications of country status, with Models 7–10 based on HDI and Models 11–14 based on per capita Gross National Income (GNI).