A linear regression model from Urban Compactness with R-squared equal one

-

What is the average travel distance in a city calculated from the compactness index? 

The compactness of an area (e.g. urban extent) is the ratio between two average distances: d / D.
There are two ways to get the numerator d:

  1. The average distance of any random point to the center of the circle, d = (128/45pi)*R = 0.9054*R
  2. The average distance of any random point to another point in the circle, d = 2/3*R (P.S., this one is easy to calculate, it is the integration of 2πr/πR^2 * r = 2r^2/R^2 from 0 to R. The idea is, for any r within range [0, R], the probability that a point on the disc is on that circle is 2πr/πR^2, then we integrate r from 0 to R. See this post.)

The circle is called the "Equal-Area circle" which has the equal area as the shape in the study (e.g. urban extent).
The denominator D is the average distance of a random point to the center or to another point in the real shape area.
  • The 1st approach (random distance to center) is called proximity compactness index;
  • The 2nd approach (between any two random points) is called cohesion compactness index; 
These two measurements are very similar and are highly correlated. They give a ratio maximized at one if the city is a perfect circle.

Compactness: How close to a circle is the shape of the city:


Figure: a city with high compactness (left, 0.86) and a city with low compactness (right, 0.49)

Here is the interesting part. When we calculated the compactness score, we also obtained (the theoretical) average travel distance in the urban extent: D. D is empirical from which we get the compactness index = d / D. If we build this linear model on the logarithmic scale:
        log_D = log_Population size + log_Population density + log_Compactness index
The R-squared almost equals one (R-squared = 0.997). Is it surprising?

Can I use linear regression to predict Y with X if Y was calculated from X?

This is a simple but sometimes confusing question.
Define: Density = Population Size / Urban Extent

  • If Y = A + B. Then using A and B as predictors for Y gives R = 1, since Y is a linear combination of A and B. 
  • If Y = A * B or Y = A / B (e.g. Density = Population / Urban Extent), a linear model of Y = b0 +  b1*A + b2*B can give a very low R-squared. Predicting density with population and urban extent directly gives R-squared = 0.03. This might be caused by the factor that population and urban extent are highly correlated (variance inflation is very high: 61), so dividing one by another might have some problems.
  • But on logarithmic scale log_Y = log_B + log_C, thus the same model on logarithmic scale gives R-squared = 1. But it doesn't mean predicting Y with either B or C alone can give strong correlation, no matter the variables are on the logarithmic scale or not. 
  • In sum, define: Density = Population Size / Urban Extent, then for these models:
    • predict Density using Population and Urban Extent                    R-squared = 0.03
    • predict log_Density using log_Population and log_Urban Extent    R-squared = 1
    • log_Population and log_Urban Extent                                             R-squared = 0.98
    • But: 
    • predict log_Density by log_Population alone                                 R-squared = 0.24
    • predict log_Density by log_Urban Extent alone                            R-squared = 0.08

*P.S. A technical notes: results above are based on Colombia Atlas, for global cities, the correlation between population and urban extent is not that high: R-squared = 0.75, and Density correlates weaker with population: R-squared = 0.08, and weaker with urban extent: R-squared =0.04.


The model for average distance

Back to the model of average distance from compactness index.
log_D = log_Population size + log_Population density + log_Compactness index
Because the numerator d = 0.91 R or d = 2/3 R. The numerator simply depends on the area of urban extent (R = radius of the equal area circle). And two cities with same population size and density have the same urban extent, thus same R, same di. Since D = d / Compactness, population size, density and Compactness fully explain (determine) the D. So we have an R-squared = 100%. It seems to be a mathematical trick, but I don't indicate it to be meaningless as it helps to show the relationship between variables from a different perspective. Here the linear model serves a more explanatory purpose than predicting.

In an earlier post that "a perfect correlation could mean nothing", we discussed correlation as a single measurement must be judged carefully in reality. The same caution could be applied to R-squared. We know in a univariate regression, R-squared is just the square of the correlation coefficient. In this case, we shall be clear that the model is for explanatory purpose and we need to understand the reason behind it. Choosing the right statistical model depends on the purpose and target. 

Extra: a nice note on the several ways to describe R-squared:

  • The strength of the relationship between variables
  • How well the predictor variables can predict the outcome
  • The amount of the variation in the response variable that is explained around its mean
  • the goodness of fit of the regression
updated 5.19

Comments

Post a Comment

Popular posts from this blog

How to Draw Heatmap with Colorful Dendrogram

-
Image
Heatmap with Dendrogram: Color Branches and Labels This data visualization example include: * Hierarchical clustering, dendrogram, and heat map based on normalized odds ratios * The dendrogram was built separately to give color to dendrogram’s branches/labels based on clusters using  dendextend * Heatmap is made by  heatmap.2  from  gplots  using the built dendrogram * The rows are sorted by means from highest to lowest, it can be done in either the dendrogram or the  heatmap.2 * Use color palettes from  colorspace Data YRBSS (Youth Risk Behavior Surveillance System) survey data from CDC. This example plots the odds ratios of 63 behavioral questions on one question regarding disordered eating. The odds ratios are median centered by column and log2 transformed Version 1: Color both the branches and labels library (curl) # read file from google drive library (gplots) # heatmap.2 library (dendextend) # make and color dendrogram library (colorspace) #
Read more

Power-law distribution (Pareto)& Zipf's Law: connection and how to fit the distribution of global city population

-
Image
- This post counts for 3 days. - In short,  We use the histogram to outline the distribution, use rank-frequency plot (log-log plot) to identify the power-law distribution, and use maximum likelihood estimation to obtain the parameters. -  I will show 1 st , how the power-law function CDF and PDF can be derived from Zipf CDF; and 2 nd , this least square estimation by excel is wrong and how to get the right one. Historical background on the power-law distribution. - Power-law distribution is as common as normal distribution in the real world. In the 19 th century, Italian economist Pareto realized that 20% of the population owns 80% the wealth, which is the famous 80/20 rule. He later named the power-law distribution as Pareto distribution, which is what we studied in the probability course. - In the 20 th century, another main contributor, Zipf made a similar discovery in linguistics about the frequency of words used. - Besides social wealth, many things follow a hig
Read more

eXtreme Gradient Boosting (XGBoost): Better than random forest or gradient boosting

-
Image
Ensemble Methods 1. Random Forest 2. Gradient boosting 3. Extreme Gradient Boosting Compared to Lasso and Best-subset Lasso Best-subset Compare MSEs Code I recently had the great pleasure to meet with Professor Allan Just and he introduced me to eXtreme Gradient Boosting (XGBoost). XGBoost is “one of the most loved machine learning algorithms at Kaggle”, it somehow combines the advantages of random forest and boosting. We will see later its predictive ability is indeed very strong. In the earlier post  Model Selection using Lasso and Best Subset  I mainly focused on model selection by best-subset, stepwise, or lasso. This post will focus on ensemble method: random forest, Gradient boosting, and eXtreme Gradient Boosting (XGBoost). An ensemble method is a technique that combines the predictions from multiple machine learning algorithms together to make more accurate predictions than any individual model. 1. Random Forest Use library  randomForest . A
Read more