Beyond the Map: Quantifying Kerala’s 2020–2025 Urban Transition

 

Over the past few weeks, many of you interacted with my earlier GIS-based analysis of Kerala’s urbanisation pattern, where I used GHS-SMOD rasters (2020 and 2025) to create village-level maps of settlement classes and transition patterns.

That study generated a lot of interest, especially around where change was happening and how strongly Kerala’s corridor-led development was reflected in the spatial data.

This new work continues from that point,
but attempts something different: by moving from visual interpretation to statistical explanation.

Why this second stage?

Maps tell us where patterns appear,
but econometrics helps us understand why they occur.

In architecture and planning, we often read patterns intuitively, by looking at maps, densities, sections, corridors, and street networks. But when we want to measure which factors actually drive those patterns, and how strongly, we need a different toolkit.

That toolkit is econometrics.

A quick, non-technical explainer: What is econometrics?

Econometrics is simply: a set of statistical methods that help us quantify relationships in real-world data.

It is not about finance or economics alone; urban researchers, transport planners, public health experts, and environmental scientists all use it.

Three broad types are especially common:

1. Cross-sectional models (like OLS): Used when we want to know how variables relate at a single point in time. E.g., Does highway proximity correlate with higher density?

2. Time-series models: Used for forecasting how something evolves over time. E.g., Predicting population or land price changes.

3. Binary or logistic models (used in this study): Used when the outcome is Yes/No. E.g., Did a village urbanise between 2020 and 2025?

Logistic models allow us to study odds; how much more likely a transition becomes when certain factors are present.

This is exactly what we need to interpret Kerala’s transition map more rigorously.

 

What this study attempts

Using 1,533 villages as spatial units, the analysis brings together:

• Raster-derived settlement classes (2020 & 2025)
• Neighbourhood urbanity (adjacency-based mean urban class)
• Accessibility distances to

a)      nearest highway

b)     nearest railway station

c)      nearest airport

• District fixed effects
• Binary and ordered transition variables derived from GIS

The central question becomes:

What factors actually drove Kerala’s village-level settlement transitions between 2020 and 2025?

 

Data Used

From QGIS, each village polygon was enriched with:

Settlement data

• SMOD 2020 class
• SMOD 2025 class
• Cleaned transition variable (up_clean: 1 = upgraded, 0 = no change/downgrade)
• Transition magnitude (d_us: ordered step-up measure)

Neighbourhood data

• Mean urbanity of touching neighbours (us20_mean)
• Count of neighbouring villages

Accessibility measures

• Distance to nearest highway
• Distance to nearest railway station
• Distance to nearest airport

All of these were exported from QGIS into a CSV, forming the basis of the R data frame.

Methods in Simple Language

Most architects and planners don’t use econometrics daily, so here is the essence:

What econometrics does here is that it tells us which factors actually drive a pattern, and how strongly.

The model used is Logistic regression, a method used when the outcome is Yes/No.
Here, the question is: What makes a village more likely to urbanise (upgrade its SMOD class) between 2020 and 2025?

The primary outcome, up_clean, is a binary indicator equal to 1 if a village upgraded its dominant SMOD class between 2020 and 2025, and 0 otherwise. For village i, the probability of upgrade is modelled using a logistic link:

where the linear predictor is:

η_i ​= β₀​ + β₁​air_HubDist_i​ + β₂​rail_HubDist_i​ + β₃​highway_HubDist_i ​+ β₄​us20_i + β₅​us20_mean_i

This baseline specification tests five theoretically motivated effects:

• Accessibility effects: closer proximity to airports and highways is expected to increase transition probability; the effect of railway distance is empirically ambiguous and treated as exploratory.
• Baseline settlement type: villages already suburban or urban in 2020 are expected to have lower upgrade probability (less room to transition).
• Neighbourhood urbanity: villages surrounded by more urban neighbours are hypothesised to exhibit higher transition probabilities due to spatial spillovers.

The baseline model provides the foundational insight into the structural correlates of transition before adjusting for district-level differences.

 

This type of model returns:

• coefficients → direction & magnitude of influence
• odds ratios → how much the probability changes
• significance levels (*, **, ***) → how confidently we can trust the effect

Model 1: Logit Results

=============================================================

                                  Dependent variable:       

                          -----------------------------------

                          Probability of Urbanisation Upgrade

-------------------------------------------------------------

Distance to Airport                   -0.00001***           

                                       (0.00000)            

                                                            

Distance to Railway                    0.00001*             

                                       (0.00001)            

                                                            

Distance to Highway                   -0.0002***            

                                       (0.00003)            

                                                            

Suburban (base=Rural)                  -3.744***            

                                        (0.271)             

                                                            

Dense (base=Rural)                     -5.320***            

                                        (0.574)             

                                                            

Urban Core (base=Rural)                 -26.406             

                                       (514.802)            

                                                            

Neighbour Urbanity (2020)              1.918***             

                                        (0.200)             

                                                            

Constant                               2.732***             

                                        (0.361)             

                                                            

-------------------------------------------------------------

Observations                             1,529              

Log Likelihood                         -601.518             

Akaike Inf. Crit.                      1,219.037            

=============================================================

Note:                             *p<0.1; **p<0.05; ***p<0.01

Table 2: Regression result of baseline Model1

How to Read This Table (Very Simply)

Most planners are not trained to interpret econometric tables, so here is the clue sheet:

• Positive value → increases the chance of urbanisation
• Negative value → decreases the chance
• *** → highly reliable
• ** → fairly reliable
• * → weakly reliable
• blank → could be noise

For example:

• “Highway distance = negative & ***”
  → the farther a village is from a highway, the much less likely it is to urbanise.
• “Neighbour urbanity = positive & ***”
  → villages surrounded by more urban villages almost always transition themselves.

These help us move from visual intuition → measurable evidence.

To make this concrete, here is what the actual model values mean:

• Highway Distance (−0.0002, *** )
  Every additional 1 km from a highway reduces the odds of urbanising by about 20%.
  Villages 5 km away have almost half the transition probability of those close to the highway.
• Airport Distance (−0.00001, *** )
  Being 10 km farther from an airport lowers the odds by roughly 10%.
• Railway Distance (+0.00001, * )
  A weak positive effect (statistically marginal) reflecting Kerala’s already railway-dense belts.
• Neighbour Urbanity (1.918, *** )
  Villages surrounded by more urban neighbours are 6–7 times more likely to urbanise themselves.
  Urbanisation spreads spatially.
• Baseline Class (compared to Rural):
  Suburban → 97% lower odds
  Dense Urban → 99% lower odds
  Urban Core → almost zero odds
  (Simply because they have little “headroom” to shift further up.)

These quantified effects transform the earlier GIS pattern from an intuitive observation into evidence-backed understanding.

What the results reveal

1. Transport accessibility matters a lot.

2. Neighbourhood effects are powerful.

3. District boundaries matter far less than we think.

Why this matters for us architects, planners, and in policy conversations

This combined GIS + econometric approach moves us from:

“It looks like a pattern” → “This is the measurable reason behind the pattern.”

For planning discussions, infrastructure prioritisation, LAP preparation, or regional mobility strategies, evidence like this helps shift debates from intuition to measurable drivers.

And for young designers or researchers, this demonstrates how spatial and statistical thinking can complement each other; visualisation tells us what is happening, econometrics tells us why.