Indian vehicle ownership and travel behavior: A case study of Bengaluru, Delhi and Kolkata

The study investigates how Indian households in three major metropolitan regions—Bengaluru, Delhi, and Kolkata—own and use motor vehicles and how these patterns may evolve. Motivated by the absence of individual- or household-level ownership models for these key cities and by the economic and environmental implications of India’s rapid motorization, the research aims to model (a) annual vehicle-kilometers traveled (VKT), (b) ownership and counts of two- and four-wheel vehicles, and (c) body-type and manufacturer-origin preferences. Understanding these determinants and forecasting future ownership levels provides insights for policymakers, industry, and urban planners concerned with congestion, emissions, safety, and infrastructure planning.

Prior Indian vehicle ownership studies have focused on regions like Mumbai, Chennai, Pune, and Surat (e.g., Kumar & Krishna, 2006; Shirgaokar et al., 2012; Gopisetty & Srinivasan, 2013; Padmini & Dhingra, 2010; Banerjee et al., 2010). Dash et al. (2013) developed an MNL model using national expenditure census data with four ownership categories (none, two-wheeler only, four-wheeler only, both), identifying expenditures, household size, children, and age as significant. Expert interviews (Bansal & Kockelman, 2017) highlighted purchase price as the dominant factor in car ownership; several experts noted that households able to afford a four-wheeler generally buy one irrespective of two-wheeler prices. Few prior studies examined preferences for vehicle body types or detailed usage (VKT), with Banerjee et al. (2010) a notable exception for body type. There was a gap in household-level modeling for Delhi, Bengaluru, and Kolkata, which this study addresses.

Data collection: A 77-question web survey (Qualtrics) was administered via Survey Sampling International’s panel during July–August 2015 in Delhi, Bengaluru, and Kolkata. Topics included current vehicle inventory (make, model, fuel, odometer), usage frequency, annual VKT per vehicle, past vehicle sales, future purchase preferences (WTP, fuel economy, body type, used vs. new, EV inclination), responses to fuel price and helmet laws, brand and operating cost importance, trip generation and mode choice, and demographics (household size, workers, children, incomes, education, home ownership, assets like cell phones, appliances, internet, credit cards).

Sample and cleaning: 1,594 completed surveys; 1,001 retained after removing speeders (<13 minutes), respondents <18 years, internal inconsistencies (e.g., vehicles owned mismatches, personal income > household income, implausible purchases), and other conflicts. Summary statistics (Tables 1–3) show higher education/income than national averages, consistent with online, English-language sampling in urban regions.

Modeling approach:

• VKT modeling: Ordinary Least Squares (OLS) regression for annual household four-wheeler VKT using demographics and location indicators. Variables included age, personal income, household size, employment, city dummies (Delhi, Kolkata), and education (bachelor’s, master’s+).
• Vehicle type ownership: Multinomial Logit (MNL) with four alternatives: no motorized vehicles (base), two-wheeler(s) only, four-wheeler(s) only, both. Estimation with R’s VGAM; iterative variable selection removing household-level attributes with p>0.2 across alternatives.
• Four-wheeler count: Poisson regression for the number of four-wheel vehicles owned/leased per household; negative binomial tested but no overdispersion detected. R’s GLM with stepwise AIC selection; retained variables significant or near-significant contributors.
• Four-wheeler body type: MNL for passenger car (base) vs truck/SUV vs van.
• Manufacturer region: MNL classifying manufacturer headquarters into India, Asia (non-India), and EU/US; explored effects of income, city, home ownership.

Forecasting framework: A static microsimulation using sampled households as the base. For each household, probability of owning four-wheelers and expected counts (Poisson model) were computed, then body type and manufacturer region were assigned probabilistically (MNLs). Exogenous inputs included GDP per capita forecasts (Wilson & Purushothaman, 2003) and cellphone ownership trajectories (eMarketer, 2014; extrapolated toward 91% per Pew, 2013) used as proxies for wealth/household asset trends. Population forecasts from United Nations (2015) informed scaling. The model produced time-paths for vehicles-per-capita and composition through 2030.

Descriptive statistics (Tables 1–3):

• Sampled households were relatively affluent and educated; mean household income ≈ INR 1,005,000; mean household size 4.14; high asset ownership (e.g., mean 3.5 cell phones, 1.75 computers).
• Annual four-wheeler household VKT averaged 8,817 km (SD 6,278), lower than the US (≈34,000 km/yr in NHTS 2009). Delhi had higher VKT than Kolkata.

VKT OLS (Table 4; dependent: annual household VKT for four-wheelers; N=880; R^2=0.064):

• Positive and significant predictors: personal income (coef 0.001145, std coef 0.139, p<0.001), age (49.15, 0.092, p=0.008), household size (166.3, 0.065, p=0.041). Kolkata residence negative (−1,838, std −0.058, p=0.001). Employment marginally positive (941, p=0.089). Education (bachelor’s or master’s+) associated with lower VKT holding other factors constant (bachelor’s: −2,360, p=0.040; master’s+: −1,677, p=0.154). Delhi residence positive but not significant (670, p=0.154).

Vehicle type MNL (Table 6; base: no motorized vehicles):

• Household income positively associated with owning two-wheeler only, four-wheeler only, and both (all p<0.001).
• More children and more married couples increase likelihoods of two-wheeler and both, and married couples also increase four-wheeler ownership.
• Residing in Kolkata significantly reduces probabilities for all motorized ownership categories (p<0.05). Residing in Delhi reduces probability of two-wheeler-only ownership (p=0.003).

Four-wheeler counts (Poisson; Table 7; N=1001; AIC=2343.2; McFadden’s R^2=0.0568):

• Positive predictors: household income (2.02e-07, p<0.001), number of credit cards (0.111, p<0.001), married couples (0.149, p=0.005); cellphones marginally positive (0.0375, p=0.078); home ownership marginally positive (0.159, p=0.054).

Body types (Tables 8–9):

• Ownership shares (n=873): cars 62%, vans 29%, trucks/SUVs 10% (vs US 2009: cars 50%, vans 8%, trucks/SUVs 37%).
• MNL: Van ownership less likely with higher household size (−0.144, p=0.014), employment (−0.484, p=0.001), and income (−3.77e-07, p=0.016). Van preference higher in Bengaluru (0.813, p=0.047) and Delhi (0.674, p=0.033). Trucks/SUVs less likely in Kolkata (−0.525, p=0.033).

Manufacturer region (Table 10 and narrative):

• Shares: India 43.5% (Maruti 35.4% largest), Asia 37.3% (Hyundai 21.6% largest), EU/US 19.1% (Ford 5.6%). Lower-income households and Kolkata residents more likely to select Indian makes; home ownership associated with higher likelihood of foreign-branded vehicles; highest-income households more likely to select EU/US brands, middle-income favor other Asian brands.

Forecasts (Table 11):

• Four-wheel vehicles per capita in regions of study: 0.238 (2015) → 0.303 (2020) → 0.419 (2025) → 0.718 (2030), implying ≈27% increase by 2020 and ≈202% by 2030 relative to 2015 (≈5.7% average annual growth).
• Composition shift: cars 59%→72%, trucks/SUVs 9%→14%, vans 32%→14% by 2030.
• Manufacturer shares shift: Indian 51%→36%, EU/US 21%→46% by 2030.

Trip generation and mode choice (Table 12):

• Average 16.9 round trips/week (≈2.42/day). Work trips comprise 4.376/wk (25.9% of weekly trips). Mode shares vary by purpose; for work: 37% drive alone 4-wheeler, 29% drive alone 2-wheeler, 8% public transport, 7% employer’s bus/taxi, 4% walk.

The study addresses key questions about determinants of vehicle ownership and usage in major Indian metros and their evolution under plausible economic and technology asset (cellphone) growth. Findings show that income, household size/structure (children, married couples), and location materially influence both vehicle ownership types and usage. Lower VKT relative to the US aligns with denser urban form, congestion, and higher reliance on transit and active modes. The negative associations of higher education with VKT (controlling for income) suggest proximity to destinations or time constraints among more educated respondents, while Kolkata’s lower ownership and VKT likely reflect transit alternatives and urban structure.

Forecasts indicate rapid motorization in the sampled regions, with strong growth in per-capita four-wheeler ownership and a shift toward cars and EU/US-branded vehicles. These trends imply rising congestion, emissions, and safety externalities unless mitigated by policy (e.g., transit investments, pricing, parking management) or technology (e.g., shared mobility, electrification). The modeling framework, though static and sample-specific, connects micro-level household attributes to macro outcomes, informing manufacturers’ market strategies and policymakers’ infrastructure and environmental planning. Discrepancies between forecasted ownership levels and administrative registrations underscore sample bias, emphasizing cautious interpretation and the need for more representative data.

This work develops and estimates an integrated set of models—OLS for VKT, MNLs for vehicle type, body type, and manufacturer region, and a Poisson count model for four-wheelers—using a new household survey of 1,001 Indians in Bengaluru, Delhi, and Kolkata. Results confirm the central role of income and household structure in ownership and use, reveal regional differences (notably lower ownership/VKT in Kolkata), and document current body-type and manufacturer preferences. A static forecasting exercise, pivoting on GDP-per-capita and cellphone ownership trajectories, projects a 202% increase in four-wheeler ownership per capita by 2030 in the sampled regions, with a growing share of cars and EU/US-branded vehicles.

Future research should employ larger, more representative samples (including non-Internet households and additional cities), enable dynamic longitudinal modeling, and incorporate richer objective travel data (e.g., GPS/smartphone traces). Policy-sensitive simulations considering pricing, transit enhancements, electrification, and shared/autonomous vehicles could refine forecasts and assess strategies to manage congestion, emissions, and safety outcomes.

The survey sample is biased toward wealthier, younger, and more educated urban residents due to online, English-language administration and panel composition, limiting generalizability beyond the sampled regions. India’s census lacks accessible cross-tabulated household microdata, precluding robust weighting or bias correction. Responses with short completion times or inconsistencies were removed, but measurement error remains possible. The VKT model has modest explanatory power (R^2≈0.064). Forecasts are static and rely on proxies (cellphone ownership) and external GDP/population projections, and they assume sampled households’ evolution reflects broader regional trends. Manufacturer origin results reflect reported brands, which may be produced domestically despite foreign design headquarters.