Interpretation of the Correlation Coefficient
The correlation coefficient as measures how strong a linear relationship exists between two numeric variables x and y. Specifically:
- The correlation coefficient is always a number between -1.0 and +1.0.
- If the correlation coefficient is close to +1.0, then there is a strong positive linear relationship between x and y. In other words, if x increases, y also increases.
- If the correlation coefficient is close to -1.0, then there is a strong negative linear relationship between x and y. In other words, if x increases, y will decrease.
- The closer to zero the correlation coefficient is, the less of a linear relationship between x and y exists
This explanation of the Correlation Coefficient is taken from:
The NYPD Stop & Frisk Summary Data Provided by the NYCLU:
The data used here is for the years 2003 to 2012. The NYCLU summary data does not include the details by race for 2002. Also, the data does not include a value for the "Age 14-25" category for 2012.
The race categories for White, Black & Latino are given in the summary data provided on the NYCLU web page. The category for "Other" is a calculated remainder:
Other = Total S&F - White - Black - Latino
It represents Asians and certain indigenous people living in New York City.
The NYCLU summary data gives a value for "Totally Innocent." "Not Innocent" is calculated as:
Not Innocent = Total S&F - Totally Innocent
Analysis of the full 2011 NYPD Stop & Frisk data set suggests that "Not Innocent" represents suspects who were arrested, given a summons or both. It is important to remember the "Not Innocent" does not mean "Guilty." We do not have data from court records to determine how the arrests and summons were adjudicated.
You can find the NYCLU summary data here:
You can find a copy of my full study explaining this data in more detail here:
Stop & Frisk Correlations to Not Innocent:
All of the race categories (except Other) show strong correlations to Not Innocent results. Correlations over +90% are very strong correlations. The Other category show moderate correlation. There is also a strong correlation to the Age 14-25 age category. Thus there is a strong linear relationship between each category and Not Innocent results. Thus there is no reason to favor one race category over another when choosing suspects to stop & frisk.
Mike Bloomberg & Ray Keller have commented that 95% of crimes are committed by young black males and 95% of those crimes are committed in black neighborhoods. The COMSTAT program helps identify which neighborhoods. Even if this is true (I have not examined crime statistics in my analysis) it is not proof that 95% of young black males are criminals! The current stop & frisk program is treating all young black males as criminals. And that is racist.
Change in Stop & Frisk Correlations to Change in Not Innocent:
The Greek letter Delta (Δ) is used in math to signify "change in" such as change in speed. The exclamation point (!) is used in some programming languages to mean negation, or "Not", thus "!Innocent" means "Not Innocent."
The notation for Blacks, "ΔB% to Δ!I%" should be interpreted to mean, "The change in Black percent over time compared to the change in Not Innocent percent over time." Interprting the results:
- Black: ΔB% to Δ!I%: The correlation is a mildly negative -11.5%. The increasing focus on Blacks over time is having negative results and is not producing a proportionate number of Not Innocent results. It is possible that Blacks are adapting thei behavior to avoid an arrest or summons during a stop & frisk. (More on this in my next post.)
- White: ΔW% to Δ!I%: The correlation is a mildly positive 3.1% suggesting that somewhat more focus on Whites might be warranted.
- Latino: ΔL% to Δ!I%: Another correlation that is a mildly positive 23.2%, also suggesting that more focus on Latinos might be warranted. This is the strongest positive for a race group.
- Other: ΔO% to Δ!I%: Another mildly negative -6.19%. The NYPD's focus on "Others" has been decreasing,over time, possibly because greater focus isn't warranted.
- Young Age: ΔYA% to Δ!I%: The Age Group 14-25 has a moderately positive correlation of 47.4%. It is the strongest positive correlation among all the categories
More on these conclusions in my following posts.