# Creating a basemap in python using contextily

Me and Gio received a peer review asking for a nice basemap in Philadelphia showing the relationship between hospital locations and main arterials for our paper on shooting fatalities.

I would typically do this in ArcMap, but since I do not have access to that software anymore, I took some time to learn the contextily library in python to accomplish the same task.

Here is the map we will be producing in the end:

So if you are a crime analyst working for a specific city, it may make sense to pull the original vector data for streets/highways and create your own style for your maps. That is quite a bit of work though, so for a more general solution these basemaps are really great. (And are honestly nicer than I could personally make even with the original vector data anyway).

Below I walk through the python code, but the data to replicate my paper with Gio can be found here, including the updated base map python script and shapefile data.

## Front matter

So first, I have consistently had a difficult time working with the various geo tools in python on my windows machine. Most recently the issue was older version of pyproj and epsg codes were giving me fits. So at the recommendation of the geopandas folks, I just created my own conda environment for geospatial stuff, and that has worked nicely so far.

So here I need geopandas, pyproj, & contexily as non-traditional libraries. Then I change my working directory to where I have my data, and then update my personal matplotlib defaults.

'''
Python script to make a basemap
'''

import geopandas
import pyproj
import contextily as cx
import matplotlib
import matplotlib.pyplot as plt
import os
os.chdir(r'D:\Dropbox\Dropbox\School_Projects\Shooting_Survival_Philly\Analysis\OriginalData'

#Plot theme
andy_theme = {'axes.grid': True,
'grid.linestyle': '--',
'legend.framealpha': 1,
'legend.facecolor': 'white',
'legend.fontsize': 14,
'legend.title_fontsize': 16,
'xtick.labelsize': 14,
'ytick.labelsize': 14,
'axes.labelsize': 16,
'axes.titlesize': 20,
'figure.dpi': 100}

matplotlib.rcParams.update(andy_theme)

## Data Prep with geopandas & pyproj

The next part we load in our shapefile into a geopandas data frame (just a border for Philly), then I just define the locations of hospitals (with level 1 trauma facilities) in text in the code.

Note that the background is in projected coordinates, so then I use some updated pyproj code to transform the lat/lon into the local projection I am using.

I thought at first you needed to only use typical web map projections to grab the tiles, but Dani Arribas-Bel has done a bunch of work to make this work for any projection. So I prefer to stick to projected maps when I can.

If you happened to want to stick to typical web map projections though geopandas makes it quite easy using geo_dat.to_crs('epsg:4326').

#####################################
#DATA PREP

ph_gp = geopandas.GeoDataFrame.from_file('City_Limits_Proj.shp')

#Locations of the hospitals
hos = [('Einstein',40.036935,-75.142657),
('Hahneman',39.957284,-75.163222),
('Temple',40.005507,-75.150257),
('Jefferson',39.949121,-75.157631),
('Penn',39.949819,-75.192883)]

#Convert to local projection
transformer = pyproj.Transformer.from_crs("epsg:4326", ph_gp.crs.to_string())
hx = []
hy = []
for h, lat, lon in hos:
xp, yp = transformer.transform(lat, lon)
hx.append(xp)
hy.append(yp)
#####################################

## Making the basemap

Now onto the good stuff. Here I use the default plotting methods from geopandas boundary plot to create a base matplotlib plot object with the Philly border outline.

Second I turn off the tick marks.

Next I have some hacky code to do the north arrow and scale bar. The north arrow is using annotations and arrows, so this just relies on the fact that north is up in the plot. (If it isn’t, you will need to adjust this for your map.)

The scale bar is more straightforward – I just plot a rectangle on the matplotlib plot, and then put text in the middle of the bar. Since the projected units are in meters, I just draw a rectangle that is 5 kilometers longways.

Then I add in the hospital locations. Note I gave the outline a label, as well as the hospitals. This is necessary to have those objects saved into the matplotlib legend. Which I add to the plot after this, and increase the default size.

Finally I add my basemap. I do not need to do anything special here, the contextily add_basemap function figures it all out for me, given that I pass in the coordinate reference system of the basemap. (You can take out the zoom level argument at first, 12 is the default zoom for Philly.)

Then I save the file to a lower res PNG.

#####################################
#Now making a basemap in contextily

ax = ph_gp.boundary.plot(color='k', linewidth=3, figsize=(12,12), label='City Boundary', edgecolor='k')
#ax.set_axis_off() #I still want a black frame around the plot
ax.get_xaxis().set_ticks([])
ax.get_yaxis().set_ticks([])

x, y, arrow_length = 0.85, 0.10, 0.07
ax.annotate('N', xy=(x, y), xytext=(x, y-arrow_length),
ha='center', va='center', fontsize=20,
xycoords=ax.transAxes)

x, y, scale_len = 829000, 62500, 5000 #arrowstyle='-'
scale_rect = matplotlib.patches.Rectangle((x,y),scale_len,200,linewidth=1,edgecolor='k',facecolor='k')
plt.text(x+scale_len/2, y+400, s='5 KM', fontsize=15, horizontalalignment='center')

plt.scatter(hx, hy, s=200, c="r", alpha=0.5, label='Trauma Hospitals')

#Now making a nice legend
ax.legend(loc='upper left', prop={'size': 20})

#Now adding in the basemap imagery

#Now exporting the map to a PNG file
plt.savefig('PhillyBasemap_LowerRes.png', dpi=100) #bbox_inches='tight'
#####################################

And voila, you have your nice basemap.

## Extra: Figuring out zoom levels

I suggest playing around with the DPI and changing the zoom levels, and changing the background tile server to see what works best given the thematic info you are superimposing on your map.

Here are some nice functions to help see the default zoom level, how many map tiles need to be downloaded when you up the default zoom level, and a list of various tile providers available. (See the contextily github page and their nice set of notebooks for some overview maps of the providers.)

#####################################
#Identifying how many tiles
latlon_outline = ph_gp.to_crs('epsg:4326').total_bounds
def_zoom = cx.tile._calculate_zoom(*latlon_outline)
print(f'Default Zoom level {def_zoom}')

cx.howmany(*latlon_outline, def_zoom, ll=True)
cx.howmany(*latlon_outline, def_zoom+1, ll=True)
cx.howmany(*latlon_outline, def_zoom+2, ll=True)

#Checking out some of the other providers and tiles
print( cx.providers.CartoDB.Voyager )
print( cx.providers.Stamen.TonerLite )
print( cx.providers.Stamen.keys() )
#####################################

# Using Association Rules to Conduct Conjunctive Analysis

I’ve suggested to folks a few times in the past that a popular analysis in CJ, called conjunctive analysis (Drawve et al., 2019; Miethe et al., 2008; Hart & Miethe, 2015), could be automated in a fashion using a popular machine learning technique called association rules. So I figured a blog post illustrating it would be good.

I was motivated by some recent work by Nix et al. (2019) examining officer involved injuries in NIBRS data. So I will be doing a relevant analysis (although not as detailed as Justin’s) to illustrate the technique.

This ended up being quite a bit of work. NIBRS is complicated, and I had to do some rewrites of finding frequent itemsets to not run out of memory. I’ve posted the python code on GitHub here. So this blog post will be just a bit of a nicer walkthrough. I also have a book chapter illustrating geospatial association rules in SPSS (Wheeler, 2017).

# A Brief Description of Conjunctive Analysis

Conjunctive analysis is more of an exploratory technique examining high cardinality categorical sets. Or in other words, you search though a database of cases that have many categories to find “interesting” patterns. It is probably easier to see an example than for me to describe it. Here is an example from Miethe et al. (2008):

You can see that here they are looking at characteristics of drug offenders, and then trying to identify particular sets of characteristics that influence the probability of a prison sentence. So this is easy to do in one dimension, it gets very difficult in multiple dimensions though.

Association rules were created for a very different type of problem – identifying common sets of items that shoppers buy together at the same time. But you can borrow that work to aid in conducting conjunctive analysis.

# Data Prep for NIBRS

So here I am using 2012 NIBRS data to conduct analysis. Like I mentioned, I was motivated by the Nix and company paper examining officer injuries. They were interested in specifically examining officer involved injuries, and whether the perception that domestic violence cases were more dangerous for officers was justified.

For brevity I only ended up examining five different variable sets in NIBRS (Justin has quite a few more in his paper):

• assault (or injury) type V4023
• victim/off relationship V4032
• ucr type V2006
• drug use V2009 (also includes computer use!)
• weapon V2017

All of these variables have three different item sets in the NIBRS codes, and many categories. You will have to dig into the python code, 00_AssocRules.py in the GitHub page to see how I recoded these variables.

Also maybe of interest I have some functions to do one-hot encoding of wide data. So a benefit of NIBRS is that you can have multiple crimes in one incident. So e.g. you can have one incident in which an assault and a burglary occurs. I do the analysis in a way that if you have common co-crimes they would pop out.

Don’t take this as very formal though. Justin’s paper which used 2016 NIBRS data only had 1 million observations, whereas here I have over 5 million (so somewhere along the way me and Justin are using different units of analysis). Also Justin’s incorporates dozens of other different variables into the analysis I don’t here.

It ends up being that with just these four variables (and the reduced sets of codes I created), there still end up being 34 different categories in the data.

# Analysis of Frequent Item Sets

The first part of conjunctive analysis (or association rules) is to identify common item sets. So the work of Hart/Miethe is always pretty vague about how you do this. Association rules has the simple approach that you find any item sets, categories in which a particular itemset meets an arbitrary threshold.

So the way you represent the data is exactly how the prior Miethe et al. (2008) data showed, you create a series of dummy 0/1 variables. Then in association rules you look for sets in which for different cases all of the dummy variables take the value of 1.

The code 01_AssocRules.py on GitHub shows this going from the already created dummy variable data. I ended up writing my own function to do this, as I kept getting out of memory errors using the mlextend library. (I don’t know if this is due to my data is large N but smaller number of columns.) You can see my freq_sets function to do this.

Typically in association rules you identify item sets that meet a particular support threshold. Support here just means the proportion of cases that those items co-occur. E.g. if 20% of cases of assault also have a weapon of fists listed. Instead though I wrote the code to have a minimum N, which I choose here to be 1000 cases. (So out of 5 million cases, this is a support of 1/5000.)

I end up finding a total of 411 frequent item sets in the data that have at least 1000 cases (out of the over 5 million). Here are a few examples, with the frequencies to the left. So there are over 2000 cases in the 2012 NIBRS data that had a known relationship between victim/offender, resulted in assault, the weapon used was fists (or kicking), and involved computer use in some way. I only end up finding two itemsets that have 5 categories and that is it, there are no higher sets of categories that have at least 1000 cases in this dataset.

3509    {'rel_Known', 'ucr_Assault', 'weap_Fists', 'ucr_Drug'}
2660    {'rel_Known', 'ucr_Assault', 'weap_Firearm', 'ucr_WeaponViol'}
2321    {'rel_Known', 'ucr_Assault', 'weap_Fists', 'drug_ComputerUse'}
1132    {'rel_Known', 'ucr_Assault', 'weap_Fists', 'weap_Knife'}
1127    {'ucr_Assault', 'weap_Firearm', 'weap_Fists', 'ucr_WeaponViol'}
1332    {'rel_Known', 'ass_Argument', 'rel_Family', 'ucr_Assault', 'weap_Fists'}
1416    {'rel_Known', 'rel_Family', 'ucr_Assault', 'weap_Fists', 'ucr_Vandalism'}

Like I said I was interested in using NIBRS because of the Nix example. One way we can then examine what variables are potentially related to officer involved injuries during a commission of a crime would be to just pull out any itemsets which include the variable of interest, here ass_LEO_Assault.

4039    {'ass_LEO_Assault'}
1232    {'rel_Known', 'ass_LEO_Assault'}
4029    {'ucr_Assault', 'ass_LEO_Assault'}
1856    {'ass_LEO_Assault', 'weap_Fists'}
1231    {'rel_Known', 'ucr_Assault', 'ass_LEO_Assault'}
1856    {'ucr_Assault', 'ass_LEO_Assault', 'weap_Fists'}

So we see there are a total of just over 4000 officer assaults in the dataset. Unsurprisingly almost all of these also had an UCR offense of assault listed (4029 out of 4039).

# Analysis of Association Rules

Sometimes just identifying the common item sets is what is of main interest in conjunctive analysis (see Hart & Miethe, 2015 for an example of examining the geographic characteristics of crime events).

But the apriori algorithm is one way to find particular rules that are of the form if A occurs then B occurs quite often, but swap out more complicated itemsets in the antecedent (A) and consequent (B) in the prior statement, and different ways of quantifying ‘quite often’.

I prefer conditional probability notation to the more typical association rule one, but for typical rules we have (here I use A for antecedent and B for consequent):

• confidence: P(A & B) / P(B). So if the itemset of just B occurs 20% of the time, and the itemset of A and B together occurs 10% of the time, the confidence would be 50%. (Or more simply the probability of B conditional on A, P(B | A)).
• lift: confidence(A,B) / P(B). This is a ratio of the baseline a category occurs for the denominator, and the numerator is the prior confidence category. So if you have a baseline B occurring 25% of the time, and the confidence of A & B is 50%, you would then have a lift of 2.

There are other rules as well that folks use, but those are the most common two I am interested in.

So for example in this data if I draw out rules that have a lift of over 2, I find rules like {'ucr_Vandalism', 'rel_Family'} -> {'ass_Argument'} produces a lift of over 6. (I can use the mlextend implementation here in this code, it was only the frequent itemsets code that was giving me problems.) So it ends up being arguments are listed in the injury codes around 1.6% of the time, but when you have a ucr crime of vandalism, and the relationship between victim/offender are family members, injury type of argument happens around 10.5% of the time (so 10.5/1.6 ~= 6).

The original use case for this is recommender systems/market analysis (so say if you see someone buy A, give them a coupon for B). So this ends up being not so interesting in this NIBRS example when you have you have more clear cause-effect type relationships criminologists would be interested in. But I describe in the next section some further potential machine learning models that may be more relevant, or how I might in the future amend the apriori algorithm for examining specific outcomes.

# Further Notes

If you have a particular outcome you are interested in a specific outcome from the get go (so not so much totally exploratory analysis as here), there are a few different options that may make more sense than association rules.

One is the RuleFit algorithm, which basically just uses a regularized regression to find simple models and low order interactions. An example of this idea using police stop data is in Goel et al. (2016). These are very similar in the end to simple decision trees, you can also have continuous covariates in the analysis and it splits them into binary above/below rules. So you could say do RTM distance analysis, and still have it output a rule if < 1000 ft predict high risk. But they are fit in a way that tend to behave better out of sample than doing simple decision trees.

Another is fitting a more complicated model, say random forests, and then having reduced form summaries to describe those models. I have some examples of using shapely values for spatial crime prediction in Wheeler & Steenbeek (2020), but for a more if-then type sets of rules you could look at Scoped Rules.

I may need to dig into the association rules code some more though, and try to update the code to take the sample sizes and statistical significance into account for a particular outcome variable. So if you find higher lift in a four set predicting a particular outcome, you search the tree for more sets with a smaller support in the distribution. (I should probably also work on some cool network viz. to look at all the different rules.)

# An intro to linear programming for criminologists

Erik Alda made the point the other day on twitter that we are one of the few crim folks that do anything related to linear programming. I think it is crazy useful – much more so than say teaching myself some new regression technique or a programming language.

I don’t quite remember the motivation to learn it. I think I kept seeing repeated applications in papers I read, but was also totally baffled by it; I did not understand peoples notation for it at all. In retrospect that was because it is not statistics. You are optimizing a function by estimating some parameters (there is nothing stochastic about it, so there is no statistical inference). So it is more like finding the min/max of a function in calculus.

I think the best way to think about linear programming is in terms of decision analysis. We have a set of options among which we need to choose some action. So we make the choices that either maximize or minimize some objective, but also take into account constraints on the decisions we can make.

For social scientists here is an example that hopefully illustrates the difference between statistics and linear programming. Say we are interested in conducting a hot spots policing randomized experiment. So we define our top 20 crime hot spots in the city, and randomly assign 10 of them to receive the hot spots treatment. Linear programming is basically the obverse of this, given our 20 hot spot areas, which are the best 10 locations to choose for our intervention.

This problem as stated you might be thinking is trivial – just rank each of the 20 hot spots by the total number of crimes, and then choose the top 10. Where linear programming really helps though is if you have constraints on the final choices you make. Say you did not want to choose hot spots that are within 1 mile of each other (to spread out the hot spot interventions throughout the city). There is no simple way to sort your hot spots to obey that constraint, but you can encode that in the linear program and have the computer solve it quite easily.

There is no shortage of ways you could expand the complexity of this example hot spot decision analysis. Say you had two different types of hot spot treatments, and that they had different efficacy in different areas (one was good for property crime, and the other was better for violent crime). You might think of this as doing two separate decision analyses, where a constraint is that an area can only be assigned one of the two interventions.

Here I will provide some code examples in python using the pulp library to illustrate some more examples using data you can see in action, as well as different ways to think about linear programming problems in practice. (Technically the examples I give are all mixed integer linear programs, as the decision variables are binary 0/1.)

# Formulating Objectives and Constraints

For this example I will be simulating data, but imagine a case you are an analyst for the IRS, and you want to determine which business tax returns to audit. We want to audit cases that both have a high probability of being fraudulent, as well as cases in which the total amount of the underpayment is large. (If you want a more typical criminology example, imagine assigning criminal cases to detectives, some cases have more costs, e.g. homicide vs burglary, and some cases have different probabilities of being solvable. This type of decision problem is very common in my experience – pretty much any time you have to make a binary choice, and those choices have variable costs/benefits.)

First I start off by simulating some data (the only libraries we need are numpy and pulp). So I simulate 1000 business tax returns, which have an estimate of the probability they are fraud, prob_fraud, and an estimate of the amount they underpayed, underpay_est.

import numpy as np
import pulp

###########################################################
#Simulate data for costs and probabilities

np.random.seed(10)
total_cases = 1000
underpay_est = np.random.uniform(1000,100000,total_cases)
prob_fraud = np.random.uniform(0,1,total_cases)
exp_return = prob_fraud*underpay_est

###########################################################

The objective we will be maximizing then is the expected return of auditing a tax return, exp_return, which is simply the multiplication of the probability of fraud multiplied by the amount of the underpayment. For a simple example, say we have a case where fraud is estimated to be 50%, and the estimate of the underpayment amount is $10,000. So our expected return for auditing that case is$5,000.

We need these two estimates external to our linear programming problem, and they themselves can be informed by predictive models (or simpler estimates, e.g. underpayment is proportional ~30% of deductions or something like that).

Now we have all we need to set up our linear programming problem. I am going to choose 100 cases out of these 1000 to audit. Hopefully that code is documented enough to see creating the decision variables (each tax return either gets a 1 if it is chosen, or a 0 if it is not), the original objective function that we are maximizing, and the results.

#Setting up the problem
case_index = list(range(total_cases))
tot_audit = 100

####################################
#Basic Problem
P = pulp.LpProblem("Choosing Cases to Audit", pulp.LpMaximize)
D = pulp.LpVariable.dicts("Decision Variable", [i for i in case_index], lowBound=0, upBound=1, cat=pulp.LpInteger)
#Objective Function
P += pulp.lpSum( D[i]*exp_return[i] for i in case_index)
#Constraint on total number of cases audited
P += pulp.lpSum( D[i] for i in case_index ) == tot_audit
#Solve the problem
P.solve()
#Get the decision variables
dec_list = [D[i].varValue for i in case_index]
dec_np = np.asarray(dec_list)
#Expected return
print( (dec_np * exp_return).sum() )
#Should be the same
print( pulp.value(P.objective) )
#Hit rate of cases
print( (dec_np * prob_fraud).sum()/tot_audit )
####################################

If you are following along in python, you will see that the total expected return is 7,287,915, and the estimated hit rate (or clearance rate) of the audits is around 0.88.

This example would be no different if we just chose the top 100 cases based on the expected return. Say that you thought the hit rate though of 88% was too low. You will choose cases that are big dollar amounts, but not necessarily a very high probability. So you may say I want my clearance rate to be over 90% overall. That is a simple constraint to add into the above model.

####################################
#Updating the problem to constrain on the hit rate
#Above a particular threshold
hit_rate = 0.9
P += pulp.lpSum( D[i]*prob_fraud[i] for i in case_index ) >= hit_rate*tot_audit
P.solve()
#Get the decision variables
dec_list = [D[i].varValue for i in case_index]
dec_np = np.asarray(dec_list)
#Expected return is slightly lower than before
print( pulp.value(P.objective) )
#Hit rate of cases
print( (dec_np * prob_fraud).sum()/tot_audit )
####################################

So now the total expected return is lower than without the constraint, 7,229,140 (so a reduction of about $60k), but our expected hit rate is just above 90%. You may be thinking, “why not just eliminate cases with a probability of lower than 90%”, and then amongst those left over select the highest expected return. That meets your constraints, but has a lower expected return than this program! Think of this program as more tit-for-tat. High expected return / lower probability audits can still be selected with this model, but you need to balance them out with some high probability cases in response to tip the scales to meet the overall hit rate objective. # Trade-Offs and the Frontier Curve So you may be thinking, ok the trade-off to get a 90% clearance was not too bad in terms of total extra taxes collected. So why not set the constraint to 95%. When you create constraints, they always lower the objective function (lower or equal to be more precise). The question then becomes quantifying that trade off. You can subsequently vary the hit rate constraint, and see how much it changes the total expected return. Here is an example of doing that, each model only takes around a second to complete. ########################################################### #Drawing the trade-off in hit rate vs expected return hit_rate = np.linspace(0.85, 0.95, 30) total_return = [] #Function to estimate the model def const_hit_rate(er, prob, tot_n, hr): c_index = range(len(er)) Prob = pulp.LpProblem("Choosing Cases to Audit", pulp.LpMaximize) Dec = pulp.LpVariable.dicts("Decision Variable", [i for i in c_index], lowBound=0, upBound=1, cat=pulp.LpInteger) Prob += pulp.lpSum( Dec[i]*er[i] for i in c_index) Prob += pulp.lpSum( Dec[i] for i in c_index ) == tot_n Prob += pulp.lpSum( Dec[i]*prob[i] for i in c_index ) >= hr*tot_n Prob.solve() dec_li = [Dec[i].varValue for i in c_index] dec_np = np.asarray(dec_li) return pulp.value(Prob.objective), dec_np for h in hit_rate: print(f'Estimating hit rate {h}') obj, dec_res = const_hit_rate(exp_return, prob_fraud, 100, h) total_return.append(obj) ########################################################### For this simulated data example, there end up being pretty severe trade-offs in the total return after you get above 91% hit rates, so from this it may not be worth the trade-off to get a much higher hit rate in practice. Just depends on how much you are willing to trade-off one for the other. There are other ways to formulate this trade off (via bi-objective functions/soft-constraints, or weighted ranking schemes), but the blog post is long enough as is! # Other Potential Applications So in terms of my work, I have examples of using linear programs to make spatial location decisions, encode fairness constraints into predictive policing, and allocate treatment assignment with network spillovers. Erik Alda and Joseph Ferrandino have conducted frontier analysis of different criminal justice organizations, which is based on estimating the frontier curve above from data (instead of a pre-specified objective function). That is about it for criminologists that I know of, but there are plenty of applications towards criminal justice topics using linear programming (or related concepts). It is most popular among operations researchers, of which Laura Albert is one of my favorites. (Criminal Justice as a field might not exist for Albert Blumstein, who was also a very influential operations researcher.) One of the things that makes this different from more traditional quantitative work in the social sciences is that again it is not statistics – we are not testing hypotheses. The contribution is simply formulating the decision problem in a tractable way that can be solved, and the drawing of the trade-offs I showed above. It is one of the ways I really like it though – unlike saying how your regression model can be used to inform decisions, this much more explicitly shows the utility of the results of those models in some practice. # Conjoint Analysis of Crime Rankings So part of my recent research mapping crime harm spots uses cost of crime estimates relevant to police departments (Wheeler & Reuter, 2020). But a limitation of this is that cost of crime estimates are always somewhat arbitrary. For a simple example, those cost estimates are based mostly on people time by the PD to respond to crimes and devote investigative resources. Many big city PDs entirely triage crimes like breaking into vehicles though. So based on PD response the cost of those crimes are basically$0 (especially if PDs have an online reporting system).

But I don’t think the public would agree with that sentiment! So in an act of cognitive dissonance with my prior post, I think asking the public is likely necessary for police to be able to ultimately serve the publics interest when doing valuations. For some ethical trade-offs (like targeting hot spots vs increasing disproportionate minority contact, Wheeler, 2019) I am not sure there is any other reasonable approach than simply getting a bunch of peoples opinions.

But that being said, I suspected that these different metrics would provide pretty similar rankings for crime severity overall. So while it is criminology 101 that official crime and normative perceptions of deviance are not a perfect 1 to 1 mapping, most folks (across time and space) have largely similar agreement on the severity of different crimes, e.g. that assault is worse than theft.

So what I did was grab some survey ranking of crime data from the original source of crime ranking that I know of, Marvin Wolfgang’s supplement to the national crime victimization survey (Wolfgang et al., 2006). I have placed all the code in this github folder to replicate. And in particular check out this Jupyter notebook with the main analysis.

# Conjoint Analysis of Crime Ranks

This analysis is often referred to as conjoint analysis. There are a bunch of different ways to conduct conjoint analysis – some ask folks to create a ranked list of items, others ask folks to choose between a list of a few items, and others ask folks to rank problems on a Likert item 1-5 scale. I would maybe guess Likert items are the most common in our field, see for example Spelman (2004) using surveys of asking people about disorder problems (and that data is available to, Taylor, 2008).

The Wolfgang survey I use here is crazy complicated, see the codebook, but in a nutshell they had an anchoring question where they assigned stealing a bike to a value of 10, and then asked folks to give a numeric score relative to that theft for a series of 24 other crime questions. Here I only analyze one version of the questionnaire, and after eliminating missing data there are still over 4,000 responses (in 1977!).

So you could do analyze those metric scores directly, but I am doing the lazy route and just doing a rank ordering (where ties are the average rank) within person. Then conjoint analysis is simply a regression predicting the rank. See the notebook for a more detailed walkthrough, so this just produces the same analysis as looking at the means of the ranks.

About the only thing I do different here than typical conjoint analysis is that I rescale the frequency weights (just changes the degrees of freedom for standard error estimates) to account for the repeated nature of the observations (e.g. I treat it like a sample of 4000 some observations, not 4000*25 observations). (I don’t worry about the survey weights here.)

To test my assertion of whether these different ranking systems will be largely in agreement, I take Jerry’s crime harm paper (Ratcliffe, 2015), which is based on sentencing guidelines, and map them as best I could to the Wolfgang questions (you could argue with me some though on those assements – and some questions don’t have any analog, like a company dumping waste). I rescaled the Wolfgang rankings to be in a range of 1-14, same as Jerry’s, instead of 1-25.

Doing a more deep dive into the Wolfgang questions, there are definately different levels in the nature of the questions you can tease out. Folks clearly take into account both harm to the victim and total damages/theft amounts. But overall the two systems are fairly correlated. So if an analyst wants to make crime harm spots now, I think it is reasonable to use one of these ranking systems, and then worry about getting the public perspective later on down the line.

The Wolfgang survey is really incredible. In this regression framework you can either adjust for other characteristics (e.g. it asks about all the usual demographics) or look at interactions (do folks who were recently victimized up their scores). So this is really just scratching the surface. I imagine if someone redid it with current data many of the metrics would be similar as well, although if I needed to do this I don’t think I would devise something as complicated as this, and would ask people to rank a smaller set of items directly.

# References

• Ratcliffe, J.H. (2015). Towards an index for harm-focused policing. Policing: A Journal of Policy and Practice, 9(2), 164-182.
• Spelman, W. (2004). Optimal targeting of incivility-reduction strategies. Journal of Quantitative Criminology, 20(1), 63-88.
• Taylor, R.B. (2008). Impacts of Specific Incivilities on Responses to Crime and Local Commitment, 1979-1994: [Atlanta, Baltimore, Chicago, Minneapolis-St. Paul, and Seattle]. https://doi.org/10.3886/ICPSR02520.v1
• Wheeler, A.P., & Reuter, S. (2020). Redrawing hot spots of crime in Dallas, Texas. https://doi.org/10.31235/osf.io/nmq8r
• Wheeler, A.P. (2019). Allocating police resources while limiting racial inequality. Justice Quarterly, Online First.
• Wolfgang, M.E., Figlio, R.M., Tracy, P.E., and Singer, S.I. (2006). National Crime Surveys: Index of Crime Severity, 1977. https://doi.org/10.3886/ICPSR08295.v1

# Notes on matplotlib and seaborn charts (python)

My current workplace is a python shop. I actually didn’t use pandas/numpy for most of my prior academic projects, but I really like pandas for data manipulation now that I know it better. I’m using python objects (lists, dictionaries, sets) inside of data frames quite a bit to do some tricky data manipulations.

I do however really miss using ggplot to make graphs. So here are my notes on using python tools to make plots, specifically the matplotlib and seaborn libraries. Here is the data/code to follow along on your own.

# some set up

First I am going to redo the data analysis for predictive recidivism I did in a prior blog post. One change is that I noticed the default random forest implementation in sci-kit was prone to overfitting the data – so one simple regularization was to either limit depth of trees, or number of samples needed to split, or the total number of samples in a final leaf. (I noticed this when I developed a simulated example xgboost did well with the defaults, but random forests did not. It happened to be becauase xgboost defaults had a way smaller number of potential splits, when using similar defaults they were pretty much the same.)

Here I just up the minimum samples per leaf to 100.

#########################################################
#set up for libraries and data I need
import pandas as pd
import os
import numpy as np
from sklearn.ensemble import RandomForestClassifier
import matplotlib
import matplotlib.pyplot as plt
import seaborn as sns

my_dir = r'C:\Users\andre\Dropbox\Documents\BLOG\matplotlib_seaborn'
os.chdir(my_dir)

#Modelling recidivism using random forests, see below for background
#https://andrewpwheeler.com/2020/01/05/balancing-false-positives/

#Preparing the variables I want
recid_prep = recid[['Recid30','CompScore.1','CompScore.2','CompScore.3',
recid_prep['Male'] = 1*(recid['sex'] == "Male")
recid_prep['Fel'] = 1*(recid['c_charge_degree'] == "F")
recid_prep['Mis'] = 1*(recid['c_charge_degree'] == "M")
recid_prep['race'] = recid['race']

#Now generating train and test set
recid_prep['Train'] = np.random.binomial(1,0.75,len(recid_prep))
recid_train = recid_prep[recid_prep['Train'] == 1]
recid_test = recid_prep[recid_prep['Train'] == 0]

#Now estimating the model
ind_vars = ['CompScore.1','CompScore.2','CompScore.3',
dep_var = 'Recid30'
rf_mod = RandomForestClassifier(n_estimators=500, random_state=10, min_samples_leaf=100)
rf_mod.fit(X = recid_train[ind_vars], y = recid_train[dep_var])

#Now applying out of sample
pred_prob = rf_mod.predict_proba(recid_test[ind_vars] )
recid_test['prob'] = pred_prob[:,1]
#########################################################

# matplotlib themes

One thing you can do is easily update the base template for matplotlib. Here are example settings I typically use, in particular making the default font sizes much larger. I also like a using a drop shadow for legends – although many consider drop shadows for data chart-junky, they actually help distinguish the legend from the background plot (a trick I learned from cartographic maps).

#########################################################
#Settings for matplotlib base

andy_theme = {'axes.grid': True,
'grid.linestyle': '--',
'legend.framealpha': 1,
'legend.facecolor': 'white',
'legend.fontsize': 14,
'legend.title_fontsize': 16,
'xtick.labelsize': 14,
'ytick.labelsize': 14,
'axes.labelsize': 16,
'axes.titlesize': 20,
'figure.dpi': 100}

print( matplotlib.rcParams )
#matplotlib.rcParams.update(andy_theme)

#print(plt.style.available)
#plt.style.use('classic')
#########################################################

I have it commented out here, but once you define your dictionary of particular style changes, then you can just run matplotlib.rcParams.update(your_dictionary) to update the base plots. You can also see the ton of options by printing matplotlib.rcParams, and there are a few different styles already available to view as well.

# creating a lift-calibration line plot

Now I am going to create a plot that I have seen several names used for – I am going to call it a calibration lift-plot. Calibration is basically “if my model predicts something will happen 5% of the time, does it actually happen 5% of the time”. I used to always do calibration charts where I binned the data, and put the predicted on the X axis, and observed on the Y (see this example). But data-robot has an alternative plot, where you superimpose those two lines that has been growing on me.

#########################################################
#Creating a calibration lift-plot for entire test set

bin_n = 30
recid_test['Bin'] = pd.qcut(recid_test['prob'], bin_n, range(bin_n) ).astype(int) + 1
recid_test['Count'] = 1

agg_bins = recid_test.groupby('Bin', as_index=False)['Recid30','prob','Count'].sum()
agg_bins['Predicted'] = agg_bins['prob']/agg_bins['Count']
agg_bins['Actual'] = agg_bins['Recid30']/agg_bins['Count']

#Now can make a nice matplotlib plot
fig, ax = plt.subplots(figsize=(6,4))
ax.plot(agg_bins['Bin'], agg_bins['Predicted'], marker='+', label='Predicted')
ax.plot(agg_bins['Bin'], agg_bins['Actual'], marker='o', markeredgecolor='w', label='Actual')
ax.set_ylabel('Probability')
ax.legend(loc='upper left')
plt.savefig('Default_mpl.png', dpi=500, bbox_inches='tight')
plt.show()
#########################################################

You can see that the model is fairly well calibrated in the test set, and that the predictions range from around 10% to 75%. It is noisy and snakes high and low, but that is expected as we don’t have a real giant test sample here (around a total of 100 observations per bin).

So this is the default matplotlib style. Here is the slight update using my above specific theme.

matplotlib.rcParams.update(andy_theme)
fig, ax = plt.subplots(figsize=(6,4))
ax.plot(agg_bins['Bin'], agg_bins['Predicted'], marker='+', label='Predicted')
ax.plot(agg_bins['Bin'], agg_bins['Actual'], marker='o', markeredgecolor='w', label='Actual')
ax.set_ylabel('Probability')
ax.legend(loc='upper left')
plt.savefig('Mytheme_mpl.png', dpi=500, bbox_inches='tight')
plt.show()

Not too different from the default, but I only have to call matplotlib.rcParams.update(andy_theme) one time and it will apply it to all my charts. So I don’t have to continually set the legend shadow, grid lines, etc.

# making a lineplot in seaborn

matplotlib is basically like base graphics in R, where if you want to superimpose a bunch of stuff you make the base plot and then add in lines() or points() etc. on top of the base. This is ok for only a few items, but if you have your data in long format, where a certain category distinguishes groups in the data, it is not very convenient.

The seaborn library provides some functions to get closer to the ggplot idea of mapping aesthetics using long data, so here is the same lineplot example. seaborn builds stuff on top of matplotlib, so it inherits the style I defined earlier. In this code snippet, first I melt the agg_bins data to long format. Then it is a similarish plot call to draw the graph.

#########################################################
#Now making the same chart in seaborn
#Easier to melt to wide data

agg_long = pd.melt(agg_bins, id_vars=['Bin'], value_vars=['Predicted','Actual'], var_name='Type', value_name='Probability')

plt.figure(figsize=(6,4))
sns.lineplot(x='Bin', y='Probability', hue='Type', style='Type', data=agg_long, dashes=False,
markers=True, markeredgecolor='w')
plt.xlabel(None)
plt.savefig('sns_lift.png', dpi=500, bbox_inches='tight')
#########################################################

By default seaborn adds in a legend title – although it is not stuffed into the actual legend title slot. (This is because they will handle multiple sub-aesthetics more gracefully I think, e.g. map color to one attribute and dash types to another.) But here I just want to get rid of it. (Similar to maps, no need to give a legend the title “legend” – should be obvious.) Also the legend did not inherit the white edge colors, so I set that as well.

#Now lets edit the legend
plt.figure(figsize=(6,4))
ax = sns.lineplot(x='Bin', y='Probability', hue='Type', style='Type', data=agg_long, dashes=False,
markers=True, markeredgecolor='w')
plt.xlabel(None)
handles, labels = ax.get_legend_handles_labels()
for i in handles:
i.set_markeredgecolor('w')
legend = ax.legend(handles=handles[1:], labels=labels[1:])
plt.savefig('sns_lift_edited_leg.png', dpi=500, bbox_inches='tight')

# making a small multiple plot

Another nicety of seaborn is that it can make small multiple plots for you. So here I conduct analysis of calibration among subsets of data for different racial categories. First I collapse the different racial subsets into an other category, then I do the same qcut, but within the different groupings. To figure that out, I do what all good programmers do, google it and adapt from a stackoverflow example.

#########################################################
#replace everyone not black/white as other
print( recid_test['race'].value_counts() )
other_group = ['Hispanic','Other','Asian','Native American']
recid_test['RaceComb'] = recid_test['race'].replace(other_group, 'Other')
print(recid_test['RaceComb'].value_counts() )

#qcut by group
bin_sub = 20
recid_test['BinRace'] = (recid_test.groupby('RaceComb',as_index=False)['prob']
).transform( lambda x: pd.qcut(x, bin_sub, labels=range(bin_sub))
).astype(int) + 1

#Now aggregate two categories, and then melt
race_bins = recid_test.groupby(['BinRace','RaceComb'], as_index=False)['Recid30','prob','Count'].sum()
race_bins['Predicted'] = race_bins['prob']/race_bins['Count']
race_bins['Actual'] = race_bins['Recid30']/race_bins['Count']
race_long = pd.melt(race_bins, id_vars=['BinRace','RaceComb'], value_vars=['Predicted','Actual'], var_name='Type', value_name='Probability')

#Now making the small multiple plot
d = {'marker': ['o','X']}
ax = sns.FacetGrid(data=race_long, col='RaceComb', hue='Type', hue_kws=d,
col_wrap=2, despine=False, height=4)
ax.map(plt.plot, 'BinRace', 'Probability', markeredgecolor="w")
ax.set_titles("{col_name}")
ax.set_xlabels("")
#plt.legend(loc="upper left")
plt.legend(bbox_to_anchor=(1.9,0.8))
plt.savefig('sns_smallmult_niceleg.png', dpi=500, bbox_inches='tight')
#########################################################

And you can see that the model is fairly well calibrated for each racial subset of the data. The other category is more volatile, but it has a smaller number of observations as well. But overall does not look too bad. (If you take out my end leaf needs 100 samples though, all of these calibration plots look really bad!)

I am having a hellishly hard time doing the map of sns.lineplot to the sub-charts, but you can just do normal matplotlib plots. When you set the legend, it defaults to the last figure that was drawn, so one way to set it where you want is to use bbox_to_anchor and just test to see where it is a good spot (can use negative arguments in this function to go to the left more). Seaborn has not nice functions to map the grid text names using formatted string substitution. And the post is long enough, you can play around yourself to see how the other options change the look of the plot.

For a few notes on various gotchas I’ve encountered so far:

• For sns.FacetGrid, you need to set the size of the plots in that call, not by instantiating plt.figure(figsize=(6,4)). (This is because it draws multiple figures.)
• When drawing elements in the chart, even for named arguments the order often matters. You often need to do things like color first before other arguments for aethetics. (I believe my problem mapping sns.lineplot to my small multiple is some inheritance problems where plt.plot does not have named arguments for x/y, but sns.lineplot does.)
• To edit the legend in a FacetGrid generated set of charts, ax returns a grid, not just one element. Since each grid inherits the same legend though, you can do handles, labels = ax[0].get_legend_handles_labels() to get the legend handles to edit if you want.

# Using pytorch to estimate group based traj models

Deep learning, tensors, pytorch. Now that I have that seo junk out of the way 🙂 – I’ve been trying to teach myself some “Deep Learning”, as it is what all of the cool kids are doing these days.

I was having a hard time though with many of the different examples. Many are for image data, and so it was hard for me to translate that to actual applications I am interested in. Many do talk about dimension reduction and reducing to hidden layers, so I thought that was similar in nature to latent class analysis, such as group-based-trajectory-modelling (GBTM).

If you aren’t familiar with GBTM, imagine a scenario in which you cluster data, and then you estimate a different regression model to predict some outcome for each subset of the clustered data. This is just a way to do that whole set-up in one go, instead of doing each part separately. It has quite a few different names – latent class analysis and mixture modelling are two common ones. The only thing really different about GBTM is that you have repeated observations – so if you follow the same person over time, they should always be assigned to the same cluster/mixture.

In short you totally can do GBTM models in deep learning libraries (as I will show), but actually most examples that I have walked through are more akin to dimension reduction of columns (so like PCA/Canonical Correlation). But the deep learning libraries are flexible enough to do the latent class analysis I want here. As far as I can tell they are basically just a nice way to estimate systems of equations (with a ton of potential parameters, and do it on the GPU if you want).

So I took it as a challenge to estimate GBTM models in a deep learning library – here pytorch. In terms of the different architectures/libraries (e.g. pytorch, tensorflow, Vowpal Wabbit) I just chose pytorch because one of my co-workers suggested pytorch was easier to learn!

I’ve posted a more detailed notebook of the code, but it worked out quite well. So first I simulated two groups of data (50 observations in each group and 11 time periods). I added a tiny bit of random noise, so this (I was hoping) should be a pretty tame problem for the machine to learn.

The code to generate a pytorch module and have the machine churn out the gradients is pretty slick (less than 30 lines total of non-comments). Many GBTM code bases make you do the analysis in wide format (so one row is an observation), but here I was able to figure out how to set it up in long data format, which makes it real easy to generalize to unbalanced data.

It took quite a few iterations to converge though, (iterations were super fast, but it is a tiny problem, so not sure how timing will generalize) and only converged when using the Adam optimizer (stochastic gradient descent converged to an answer with a similar mean square error, but not to anywhere near the right answer). These models are notorious for converging to sub-optimal locations, so that may just be an intrinsic part of the problem and a good library needs to do better with starting conditions.

I have a few notes about potential updates to the code at the end of my Jupyter notebook. For count or binomial 0/1 data, that should be a pretty easy update. Also need to write code to do out of sample predictions (which I think I can figure out as well). A harder problem I am not sure how to figure out is to do an equation for the latent groups inside of the function. And I don’t know how to get standard errors for the coefficient estimates. Hopefully I can figure that out while trying to teach myself some more deep learning. I have a few convolution ideas I want to try out for spatial-temporal crime forecasting and include proactive police feedback, but I won’t get around to them for quite awhile I imagine.

# Balancing False Positives

One area of prediction in criminal justice I think has alot of promise is using predictive algorithms in place of bail decisions. So using a predictive instrument to determine whether someone is detained pre-trial based on risk, or released on recognizance if you are low risk. Risk can be either defined as based on future dangerousness or flight risk. This cuts out the middle man of bail, which doesn’t have much evidence of effectiveness, and has negative externalities of placing economic burdens on folks we really don’t want to pile that onto. It is also the case algorithms can likely do quite a bit better than judges in figuring out future risk. So an area I think they can really do good compared to current status quo in the CJ system.

A reasonable critique of such systems though is they can have disparate racial impact. For example, ProPublica had an article on how the Compas risk assessment instrument resulted in more false positives for black than white individuals. Chris Stucchio has a nice breakdown for why this occurs, which is not due to the Compas being intrinsically racist algorithm, but due to the nature of the baseline risks for the two groups.

Consider a very simple example to illustrate. Imagine based on our cost-benefit analysis, we determine the probability threshold to flag a individual as high risk is 60%. Now say our once we apply our predictions, for those above the threshold, whites are all predicted to be 90%, and blacks are all 70%. If our model is well calibrated (which is typically the case), the false positive rate for whites will be 10%, and will be 30% for blacks.

It is actually a pretty trivial problem though to balance false positive rates between different groups, if that is what you want to do. So I figured I would illustrate here using the same ProPublica data. There are trade-offs though with this, balancing false positives means you lose out on other metrics of fairness. In particular, it means you don’t have equality of treatment – different racial groups will have different thresholds. The full data and code I use to illustrate this can be downloaded here.

## An Example in Python

To illustrate how we would balance the false positive rates between groups, I use the same ProPublica risk assessment data. So this isn’t per se for bail decisions, but works fine as an illustration. First in python I load my libraries, and then read in the data – it is a few over 11,000 cases.

import pandas as pd
import os
import numpy as np
from sklearn.ensemble import RandomForestClassifier
import matplotlib.pyplot as plt

my_dir = r'C:\Users\andre\Dropbox\Documents\BLOG\BalanceFalsePos'
os.chdir(my_dir)

#For notes on data source, check out
#https://github.com/apwheele/ResearchDesign/tree/master/Week11_MachineLearning
print( recid.head() )

Next I prepare the dataset for modelling. I am not using all of the variables in the dataset. What I predict here is recidivism post 30 days (there are a bunch of recidivism right away in the dataset, so I am not 100% sure those are prior to screening). I use the three different aggregate compas scores, juvenile felony count, whether they were male, how old they were, and whether the current charge to precipitate screening is a felony or misdemeanor. I include the race variable in the dataset, but I won’t be using it in the predictive model. (That point deserves another blog post, contra to what you might expect, leaving race flags in will often result in better outcomes for that protected class.)

#Preparing the variables I want
recid_prep = recid[['Recid30','CompScore.1','CompScore.2','CompScore.3',
recid_prep['Male'] = 1*(recid['sex'] == "Male")
recid_prep['Fel'] = 1*(recid['c_charge_degree'] == "F")
recid_prep['Mis'] = 1*(recid['c_charge_degree'] == "M")
recid_prep['race'] = recid['race']
print( recid['race'].value_counts() ) #pretty good sample size for both whites/blacks

Next I make my testing and training sets of data. In practice I can perfectly balance false positives retrospectively. But having a test set is a better representation of reality, where you need to make some decisions on the historical data and apply it forward.

#Now generating train and test set
recid_prep['Train'] = np.random.binomial(1,0.75,len(recid_prep))
recid_train = recid_prep[recid_prep['Train'] == 1]
recid_test = recid_prep[recid_prep['Train'] == 0]

Now the procedure I suggest to balance false-positives doesn’t matter how you generate the predictions, just that we need a predicted probability. Here I use random forests, but you could use whatever machine learning or logistic regression model you want. Second part just generates the predicted probabilities for the training dataset.

#Now estimating the model
ind_vars = ['CompScore.1','CompScore.2','CompScore.3',
dep_var = 'Recid30'
rf_mod = RandomForestClassifier(n_estimators=500, random_state=10)
rf_mod.fit(X = recid_train[ind_vars], y = recid_train[dep_var])

#Now getting the predicted probabilities in the training set
pred_prob = rf_mod.predict_proba(recid_train[ind_vars] )
recid_train['prob'] = pred_prob[:,1]
recid_train['prob_min'] = pred_prob[:,0]

Now to balance false positives, I will show a graph. Basically this just sorts the predicted probabilities in descending order for each racial group. Then you can calculate a cumulate false positive rate for different thresholds for each group.

#Making a cusum plot within each racial group for the false positives
recid_train.sort_values(by=['race','prob'], ascending=False, inplace=True)
recid_train['const'] = 1
recid_train['cum_fp'] = recid_train.groupby(['race'])['prob_min'].cumsum()
recid_train['cum_n'] = recid_train.groupby(['race'])['const'].cumsum()
recid_train['cum_fpm'] = recid_train['cum_fp'] / recid_train['cum_n']
white_rt = recid_train[recid_train['race'] == 'Caucasian']
black_rt = recid_train[recid_train['race'] == 'African-American' ] 

And now the fun part (and least in output, not really in writing matplotlib code).

#now make the chart for white and black
fig, ax = plt.subplots()
ax.plot(black_rt['prob'], black_rt['cum_fpm'], drawstyle='steps', color='b', label='Black')
ax.plot(white_rt['prob'], white_rt['cum_fpm'], drawstyle='steps', color='r', label='White')
ax.set_xlim(1, 0)  # decreasing probs
plt.xticks(np.arange(1.0,-0.1,-0.1))
ax.set_xlabel('Predicted Probability')
ax.set_ylabel('Mean False Positive Rate')
ax.grid(True,linestyle='--')
ax.legend(facecolor='white', framealpha=1)
plt.savefig('FP_Rate.png', dpi=2000, bbox_inches='tight')
plt.show()

So what this chart shows is that if we set our threshold to a particular predicted probability (X axis), based on the data we would expect a false positive rate (Y axis). Hence if we want to balance false positives, we just figure out the race specific thresholds for each group at a particular Y axis value. Here we can see the white line is actually higher than the black line, so this is reverse ProPublica findings, we would expect whites to have a higher false positive rate than blacks given a consistent predicted probability of high risk threshold. So say we set the threshold at 10% to flag as high risk, we would guess the false positive rate among blacks in this sample should be around 40%, but will be closer to 45% in the white sample.

Technically the lines can cross at one or multiple places, and those are places where you get equality of treatment and equality of outcome. It doesn’t make sense to use that though from a safety standpoint – those crossings can happen at a predicted probability of 99% (so too many false negatives) or 0.1% (too many false positives). So say we wanted to equalize false positive rates at 30% for each group. Here this results in a threshold for whites as high risk of 0.256, and for blacks a threshold of 0.22.

#Figuring out where the threshold is to limit the mean FP rate to 0.3
#For each racial group
white_thresh = white_rt[white_rt['cum_fpm'] > 0.3]['prob'].max()
black_thresh = black_rt[black_rt['cum_fpm'] > 0.3]['prob'].max()
print( white_thresh, black_thresh )

Now for the real test, lets see if my advice actually worked in a new sample of data to balance the false positive rate.

#Now applying out of sample, lets see if this works
pred_prob = rf_mod.predict_proba(recid_test[ind_vars] )
recid_test['prob'] = pred_prob[:,1]
recid_test['prob_min'] = pred_prob[:,0]

white_test = recid_test[recid_test['race'] == 'Caucasian']
black_test = recid_test[recid_test['race'] == 'African-American' ]

white_test['Flag'] = 1*(white_test['prob'] > white_thresh)
black_test['Flag'] = 1*(black_test['prob'] > black_thresh)

white_fp= 1 - white_test[white_test['Flag'] == 1][dep_var].mean()
black_fp = 1 - black_test[black_test['Flag'] == 1][dep_var].mean()
print( white_fp, black_fp )

And we get a false positive rate of 54% for whites (294/547 false positives), and 42% for blacks (411/986) – yikes (since I wanted a 30% FPR). As typical, when applying your model to out of sample data, your predictions are too optimistic. I need to do some more investigation, but I think a better way to get error bars on such thresholds is to do some k-fold metrics and take the worst case scenario, but I need to investigate that some more. The sample sizes here are decent, but there will ultimately be some noise when deploying this in practice. So basically if you see in practice the false positive rates are within a few percentage points that is about as good as you can get in practice I imagine. (And for smaller sample sizes will be more volatile.)

# Optimal treatment assignment with network spillovers

Motivated by a recent piece by Wood and Papachristos (2019), (WP from here on) which finds if you treat an individual at high risk for gun shot victimization, they have positive spillover effects on individuals they are connected to. This creates a tricky problem in identifying the best individuals to intervene with given finite resources. This is because you may not want to just choose the people with the highest risk – the best bang for your buck will be folks who are some function of high risk and connected to others with high risk (as well as those in areas of the network not already treated).

For a simplified example consider the network below, with individuals baseline probabilities of future risk noted in the nodes. Lets say the local treatment effect reduces the probability to 0, and the spillover effect reduces the probability by half, and you can only treat 1 node. Who do you treat?

We could select the person with the highest baseline probability (B), and the reduced effect ends up being 0.5(B) + 0.1(E) = 0.6 (the 0.1 is for the spillover effect for E). We could choose node A, which is a higher baseline probability and has the most connections, and the reduced effect is 0.4(A) + 0.05(C) + 0.05(D) + 0.1(E) = 0.6. But it ends up in this network the optimal node to choose is E, because the spillovers to A and B justify choosing a lower probability individual, 0.2(E) + 0.2(A) + 0.25(B) = 0.65.

Using this idea of a local effect and a spillover effect, I formulated an integer linear program with the same idea of a local treatment effect and a spillover effect:

$\text{Maximize} \{ \sum_{i = 1}^n (L_i\cdot p_{li} + S_i \cdot p_{si}) \}$

Where $p_{li}$ is the reduction in the probability due to the local effect, and $p_{si}$ is the reduction in the probability due to the spillover effect. These probabilities are fixed values you know at the onset, e.g. estimated from some model like in Wheeler, Worden, and Silver (2019) (and Papachristos has related work using the network itself to estimate risk). Each node, i, then gets two decision variables; $L_i$ will equal 1 if that node is treated, and $S_i$ will equal 1 if the node gets a spillover effect (depending on who is treated). Actually the findings in WP show that these effects are not additive (you don’t get extra effects if you are treated and your neighbors are treated, or if you have multiple neighbors treated), and this makes it easier to keep the problem on the probability scale. So we then have our constraints:

1. $L_i , S_i \in \{ 0,1 \}$
2. $\sum L_i = K$
3. $S_i \leq 1 + -1\cdot L_i , \forall \text{ Node}_i$
4. $\sum_{\text{neigh}(i)} L_j \geq S_i , \forall \text{ Node}_i$

Constraint 1 is that these are binary 0/1 decision variables. Constraint 2 is we limit the number of people treated to K (a value that we choose). Constraint 3 ensures that if a local decision variable is set to 1, then the spillover variable has to be set to 0. If the local is 0, it can be either 0 or 1. Constraint 4 looks at the neighbor relations. For Node i, if any of its neighbors local treated decision variable is set to 1, the Spillover decision variable can be set to 1.

So in the end, if the number of nodes is n, we have 2*n decision variables and 2*n + 1 constraints, I find it easier just to look at code sometimes, so here is this simple network and problem formulated in python using networkx and pulp. (Here is a full file of the code and data used in this post.) (Update: I swear I’ve edited this inline code snippet multiple times, if it does not appear I have coded constraints 3 & 4, check out the above linked code file. Maybe it is causing problems being rendered.)

####################################################
import pulp
import networkx

Nodes = ['a','b','c','d','e']
Edges = [('a','c'),
('a','d'),
('a','e'),
('b','e')]

p_l = {'a': 0.4, 'b': 0.5, 'c': 0.1, 'd': 0.1,'e': 0.2}
p_s = {'a': 0.2, 'b': 0.25, 'c': 0.05, 'd': 0.05,'e': 0.1}
K = 1

G = networkx.Graph()

P = pulp.LpProblem("Choosing Network Intervention", pulp.LpMaximize)
L = pulp.LpVariable.dicts("Treated Units", [i for i in Nodes], lowBound=0, upBound=1, cat=pulp.LpInteger)
S = pulp.LpVariable.dicts("Spillover Units", [i for i in Nodes], lowBound=0, upBound=1, cat=pulp.LpInteger)

P += pulp.lpSum( p_l[i]*L[i] + p_s[i]*S[i] for i in Nodes)
P += pulp.lpSum( L[i] for i in Nodes ) == K

for i in Nodes:
P += pulp.lpSum( S[i] ) <= 1 + -1*L[i]
ne = G.neighbors(i)
P += pulp.lpSum( L[j] for j in ne ) >= S[i]

P.solve()

#Should select e for local, and a & b for spillover
print(pulp.value(P.objective))
print(pulp.LpStatus[P.status])

for n in Nodes:
print([n,L[n].varValue,S[n].varValue])
####################################################

And this returns the correct results, that node E is chosen in this example, and A and B have the spillover effects. In the linked code I provided a nicer function to just pipe in your network, your two probability reduction estimates, and the number of treated units, and it will pipe out the results for you.

For an example with a larger network for just proof of concept, I conducted the same analysis, choosing 20 people to treat in a network of 311 nodes I pulled from Rostami and Mondani (2015). I simulated some baseline probabilities to pipe in, and made it so the local treatment effect was a 50% reduction in the probability, and a spillover effect was a 20% reduction. Here red squares are treated, pink circles are the spill-over, and non-treated are grey circles. It did not always choose the locally highest probability (largest nodes), but did tend to choose highly connected folks also with a high probability (but also chose some isolate nodes with a high probability as well).

This problem is solved in an instant. And I think out of the box this will work for even large networks of say over 100,000 nodes (I have let CPLEX churn on problems with near half a million decision variables on my desktop overnight). I need to check myself to make 100% sure though. A simple way to make the problem smaller if needed though is to conduct the analysis on subsets of connected components, and then shuffle the results back together.

Looking at the results, it is very similar to my choosing representatives work (Wheeler et al., 2019), and I think you could get similar results with just piping in 1’s for each of the local and spillover probabilities. One of the things I want to work on going forward though is treatment non-compliance. So if we are talking about giving some of these folks social services, they don’t always take up your offer (this is a problem in choose rep’s for call ins as well). WP actually relied on this to draw control nodes in their analysis. I thought for a bit the problem with treatment non-compliance in this setting was intractable, but another paper on a totally different topic (Bogle et al., 2019) has given me some recent hope that it can be solved.

This same idea is also is related to hot spots policing (think spatial diffusion of benefits). And I have some ideas about that to work on in the future as well (e.g. how wide of net to cast when doing hot spots interventions given geographical constraints).

# References

• Bogle, J., Bhatia, N., Ghobadi, M., Menache, I., Bjørner, N., Valadarsky, A., & Schapira, M. (2019). TEAVAR: striking the right utilization-availability balance in WAN traffic engineering. In Proceedings of the ACM Special Interest Group on Data Communication (pp. 29-43).
• Rostami, A., & Mondani, H. (2015). The complexity of crime network data: A case study of its consequences for crime control and the study of networks. PloS ONE, 10(3), e0119309.
• Wheeler, A. P., McLean, S. J., Becker, K. J., & Worden, R. E. (2019). Choosing Representatives to Deliver the Message in a Group Violence Intervention. Justice Evaluation Journal, Online First.
• Wheeler, A. P., Worden, R. E., & Silver, J. R. (2019). The Accuracy of the Violent Offender Identification Directive Tool to Predict Future Gun Violence. Criminal Justice and Behavior, 46(5), 770-788.
• Wood, G., & Papachristos, A. V. (2019). Reducing gunshot victimization in high-risk social networks through direct and spillover effects. Nature Human Behaviour, 1-7.

# Finding the dominant set in a network (python)

My paper, Choosing representatives to deliver the message in a group violence intervention, is now published online at the Justice Evaluation Journal. For those who don’t have access to that journal, here is a link good for 50 e-prints (for a limited time), and here is a pre-print version, and you can always send me an email for the published copy.

I’ve posted Python code to replicate the analysis, including the original network nodes and edges group data. I figured I would go through a quick example of applying the code for others to use the algorithm.

The main idea is that for a focused deterrence initiative, for the call-ins you want to identify folks to spread the deterrence message around the network. When working with several PDs I figured looking at who was called in would be interesting. Literally the first network graph I drew was below on the left — folks who were called in are the big red squares. This was one of the main problem gangs, and the PD had done several call-ins for over a year at this point. Those are not quite the worst set of four folks to call-in based on the topology of the network, but damn close.

But to criticize the PD I need to come up with a better solution — which is the graph on the right hand side. The larger red squares are my suggested call-ins, and they reach everyone within one step. That means everyone is at most just one link away from someone who attended the call-in. This is called a dominant set of a graph when all of the graph is colored in.

Below I give a quicker example using my code for others to generate the dominant set (instead of going through all of the replication analysis). If you are a PD interested in applying this for your focused deterrence initiative let me know!

So first to set up your python code, I import all of the needed libraries (only non-standard is networkx). Then I import my set of functions, named MyFunctions.py, and then change the working directory.

############################################################
#The libraries I need

import itertools
import networkx as nx
import csv
import sys
import os

#Now importing my own functions I made
locDir = r'C:\Users\axw161530\Dropbox\Documents\BLOG\DominantSet_Python'
sys.path.append(locDir)
from MyFunctions import *

#setting the working directory to this location
os.chdir(locDir)
#print(os.getcwd())
############################################################

The next part I read in the CSV data for City 4 Gang 1, both the nodes and the edges. Then I create a networkx graph simply based on the edges. Technically I do not use the node information at all for this, just the edges that list a source and a target.

############################################################
#Reading in the csv files that have the nodes and the edges
#And turning into a networkX graph

#simple function to read in csv files
tup = []
with open(loc) as f:
for row in z:
tup.append(tuple(row))
return tup

#Turning my csv files into networkx objects

#Turning my csv files into networkx objects
C1G4 = nx.Graph()
############################################################

Now it is quite simple, to get my suggested dominant set it is simple as this function call:

ds_C1G4 = domSet_Whe(C1G4)
print(ds_C1G4)

In my current session this gives the edges ['21', '18', '17', '16', '3', '22', '20', '6']. Which if you look to my original graph is somewhat different, but all are essentially single swaps where the best node to choose is arbitrary.

I have a bunch of other functions in the analysis, one of interest will be given who is under probation/parole who are the best people to call in (see the domSet_WheSub function). Again if you are interested in pursuing this further always feel free to reach out to me.

CiteULike, an online bibliography manager, is unfortunately shutting down. They have a service to export your bibliography as a BibTex file, but this does not include the PDFs you have uploaded to the site. Having web access to the PDFs is one of the main reasons I liked CiteULike (along with the tag cloud).

I have too many PDFs to download them all manually (over 2,000), so I wrote a script in Python to download the PDFs. Unlike prior scraping examples I’ve written about, you need to have signed into your CiteULike account to be able to download the files. Hence I use the selenium library to mimic what you do normally in a web-browser.

So let me know what bibliography manager I should switch to. Really one of the main factors will be if I can automate the conversion, including PDFs (even if it just means pointing to where the PDF is stored on my local machine).

This is a good tutorial to know about even if you don’t have anything to do with CiteULike. There are various web services that you need to sign in or mimic the browser like this to download data repeatedly, such as if a PD has a system where you need to input a set of dates to get back crime incidents (and limit the number returned, so you need to do it repeatedly to get a full sample). The selenium library can be used in a similar fashion to this tutorial in that circumstance.