Data and Me

Association Rules help in discovering relationships between items (or sets of items) of a transactional database. The goal behind mining these relationships could be to understand customers’ purchasing or browsing behavior, and/or to identify baskets of items purchased together frequently. This is also often referred to as Market Basket Analysis.

For example, let us imagine a Game of Thrones transactional database with 13 customers and their transactions as recorded below:

Our goal is to find out how likely is it for a new customer ‘Joffrey’, who bought a Dishwasher, to also buy a Washer Dryer.

Sessionization: Sometimes, a transaction is not complete in just one attempt. More often, customers come back for accessories, relevant products or replacements. To capture each of these operations as one transaction, sessions can be created. A session can be reasoned to be an extended time period which may account for additional actions of purchase that could be considered as one transaction.

Let us consider 7 days as a session here. SQL-MR on Aster offers to sessionize using this sessionizing function:

In this case, a customer’s activity in one session counts as a transaction. Example, customer ‘Jon’ had purchased Decking and Paint on 15/11/2015. In six days he returned to purchase Decking Installation. These purchases, occurring in one session, are counted as one transaction. The same is not the case with Arya and Daenerys who made purchases well after 7 days following their previous purchase. Customer ID and Session ID together denote a transaction, and therefore form a Transaction ID.

Sessionized transactions help us capture near purchase events, which we might otherwise miss, while mining relationships.

Rule mining:

Once we have a set of transactions ready, either sessionized or un-sessionized, we can mine relationships from those transactions.  SQL-MR Basket Generator Function on Asters helps to create item-set pairs from transactions.

While mining rules, metrics like Support, Confidence, and Lift help determine the strength of a rule. To calculate these metrics, it is helpful to have some of the following aggregations handy.

Note: If you wish to specify a threshold or upper limit on the number of transactions that each item should have to enter the model, a ‘having clause’ can be used after grouping.

Once we have all the aggregations handy, we can make the final calculations for Rule Generation with all the required metrics.

Here’s a subset of rules (for dishwashers) that have been generated for the GOT transactional dataset:

1. INTRODUCTION

Alcoholism is a broad term for problems with alcohol consumption. It is generally used to mean compulsive and uncontrolled consumption of alcoholic beverages, usually to the disadvantage of the drinker's health, personal relationships, and social standing. Several other terms are used to describe alcoholism; specifically terms like "alcohol abuse" and "alcohol dependence”.

Not all alcohol consumers are alcohol dependent. Millions of people drink alcoholic beverages on a regular basis without ever developing a drinking problem. According to extensive research by the National Institute on Alcohol Abuse and Alcoholism (NIAAA), less that 2% of drinkers who fall within the specified level of alcohol consumption ever develop alcohol use disorders.

A lot of study has been conducted on alcohol dependence and its unfavorable effects, mostly concentrated on the associations between alcohol consumption and epidemiology, neuroscience, behavioral patterns and other drug abuse. However, less has been known about the association between Alcohol Dependence and Anti-Social Behavior. Anti-social behavior is that which lacks consideration for others and may cause damage to the society, whether intentionally or through negligence. Anti-social behavior is labeled as such when it is deemed contrary to prevailing norms for social conduct. This encompasses a large spectrum of actions. For this analysis, we shall take into account this actions or variables as listed in the data. They are:

Problems with neighbors, friends or relatives

Bankruptcy, financial crisis or inability to pay bills on time

Trouble with police, arrests and jail

There are 2 research questions or hypotheses I propose in this study.

What is approximate level of alcohol consumption that makes a person Alcohol Dependent?

Is there a significant association between Alcohol Dependence and Anti-Social Behavior?

  2. DATA

2.1. Sample Data

For this study, the sample that I used is the NESARC - National Epidemiologic Survey on Alcohol and Related Conditions. It is sponsored by U.S. Department of Health and Human Services (DHHS), National Institutes of Health and National Institute on Alcohol Abuse and Alcoholism (NIAAA).

The National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) was designed to determine the magnitude of alcohol use disorders and their associated disabilities in the general population and in subgroups of the population and to examine changes over time in alcohol use disorders and their associated disabilities. It is a longitudinal survey with its first wave of interviews fielded in 2001-2002 and second wave in 2004-2005. The NESARC data is a representative sample of the U.S. population with 43093 observations 18 years of age and older.

The data has been collected through computer-assisted personal interviews. The NESARC data used a three stage sampling design. The stage one was the primary sampling unit selection from the Census, the second stage was to select a house hold from that unit and the third stage was to select an individual from the household. Data was collected on key policy issues including alcohol and substance use as well as health.

2.2. Measures

The main variables used for this analysis are:

S2AQ8A="How often did you drink any alcohol in last 12 months (per year)"

S2AQ8B="Number of drinks of any alcohol usually consumed on days when drank alcohol in last 12 months (per day)"

ALCABDEP12DX="Alcohol abuse/dependence in last 12 months"

S1Q239="Had a problem with neighbor, friend, or relative in last 12 months"

S1Q2310="Experienced a major financial crisis, bankruptcy, or been unable to pay bills on time in last 12 months"

S1Q2311="Did you or your family member have a trouble with police, got arrested, or sent to jail in last 12 months"

*The remaining can be found in the program attached (APPENDIX – SAS Program)*

2.3. Data Management

For the purpose of the analysis, the data has been managed by setting unknown observations as missing and by creating new variables to make the data more expressive.

*Managed variables can be found in the program attached (APPENDIX – SAS Program)*

2.4. Analysis Procedures

SAS statistical software has been used for the statistical analysis of the data. Univariate and Bivariate statistical analyses were performed on the data to answer relevant research questions.

  3. RESULTS

3.1. Univariate

Univariate analysis is carried out with the description of a single variable and its attributes of the applicable unit of analysis. Most of the univariate analysis has been performed around the following 6 variables.

The frequency (number of days) of consumption of alcohol per year by the observation in the past year.

The total number of drinks consumed per year by the observation in the past year.

An observation’s alcohol dependence in the past year.

If an observation had a problem with his neighbors, friends or relatives in the past year.

If an observation experienced a bankruptcy, a financial crisis or had been unable to pay bills in the past year.

If an observation had a problem with the police, got arrested or had been sent to jail in the past year.

For the first variable i.e ‘the frequency (number of days) of consumption of alcohol per year by the observation in the past year’, the univariate analysis shows that the average number of days per year that an observation consumed alcohol is around 50.

For the second variable i.e ‘the total number of drinks consumed per year by the observation in the past year’, the average number of drinks consumed per year is around 158.

The remaining variables being categorical, five number summaries cannot be computed. However, the frequency distributions are as follows:

Though majority of the sample are not alcohol dependent and respond negatively to the anti-social behavior question, it will be interesting to understand the association between these variables for those observations with a positive response. 

3.2. Bivariate

Bivariate Analysis is the analysis of two variables for the purpose of determining the empirical relationship between them. In order to see if the variables are related to one another, it is measured how the two variables simultaneously change together.

Progressing to the research questions at hand, a bivariate analysis is required to find the associations between 2 variables.

Question 1: What is approximate level of alcohol consumption that makes a person Alcohol Dependent?

To answer this question, one has to determine if there is an association between the number of alcohol drinks consumed per year and the alcohol dependence of a person. For such purpose, a null hypothesis can be made:

Null Hypothesis: There is no significant association between the number of alcohol drinks consumed per year and the alcohol dependence of a person.

To understand the association between level of consumption and alcohol dependence, a Chi Square Test can be used. A Chi Square test is being used as opposed to other techniques like ANOVA or Correlation because both the explanatory and the response variables are categorical.

The Chi Square value is very high i.e 7920.5920 with a p value <0.0001. In statistical significance testing the p-value is the probability of obtaining a test statistic at least as extreme as the one that was actually observed, assuming that the null hypothesis is true. Null hypothesis is rejected when the p-value is less than the predetermined significance level which is often 0.05, indicating that the observed result would be highly unlikely under the null hypothesis.

If the categorical explanatory variable has more than 2 categories, the Bonferroni Adjustment is applied to control the family wise error rate. The categorical explanatory variable here, Consumption Category has 6 categories and thus has 15 comparisons. The Bonferroni adjusted p value is 0.05/15=0.003. The Null hypothesis will be rejected only if the p value is less than 0.003.

  According to the above figure, the Bonferroni adjusted p value is less than 0.003 for almost all values except between the category comparisons 3-5 and 4-5. This could be due to a simple reason that the small number of observations that fall in category 5 is not enough to make a comparison significant. There are only 5 observations in category 5. However, the overall test is significant.

Furthermore, from the figure below, it can be inferred that for consumption categories 3 and above, the rate of alcohol dependence has increased to more than 50%.

Keeping this evidence in mind, the Null Hypothesis that there is no significant association between the number of alcohol drinks consumed per year and the alcohol dependence of a person can be rejected.

However, to determine the approximate level of alcohol that makes a person alcohol dependent, the below figure can be studied again. To summarize the figure, the rate of alcohol dependence seems to be increasing with increasing consumption category. For the observations who abstained from alcohol in the past year, the alcohol dependence is negligible. The rate of alcohol dependence is still very low for those who consumed less than 500 drinks per year– 7.28%. The rate of alcohol dependence slowly increased for those who consumed 500 drinks and above per year. However, for those who consumed 1000 drinks and above per year, the rate of alcohol dependence has increased to more than 50% of the observations. However, a primary conclusion can be drawn that the approximate level of alcohol that makes a person alcohol dependent is around 1000 drinks and above per year. 

  Question 2: Is there a significant association between Alcohol Dependence and Anti-Social Behavior?

The Null hypothesis can be stated as

Null Hypothesis: There is no significant association between Alcohol Dependence and Anti-Social Behavior.

To understand the association between these two variables, a Chi Square test can be used. No Bonferroni adjustment is required as the explanatory and response variables have only 2 categories each.

For this analysis, we shall take into account the following variables as Anti-Social Behavior.

Problems with neighbors, friends or relatives

Bankruptcy, financial crisis or inability to pay bills on time

Trouble with police, arrests and jail

Therefore, the Null Hypothesis can be further divided into three hypotheses. 

a) There is no significant association between Alcohol Dependence and Having a problem with a friend, neighbor or relative

The Chi Square value for the above test is 288.98 with a p value < 0.0001. The Null hypothesis can be rejected.

b) There is no significant association between Alcohol Dependence and Experiencing a major financial crisis, going bankrupt or being unable to pay bills on time

The Chi Square value for the test is 282.35 with a p value <0.0001. The Null hypothesis can be rejected.

c) There is no significant association between Alcohol Dependence and Having a trouble with Police, getting arrested or getting sent to jail

The Chi Square value for this test is 889.49 with a p value <0.0001. The Null hypothesis can be rejected.

  3.3. Moderation

To find out if the significant association between alcohol dependence and anti-social behavior is unadultered, the relation is moderated with another variable which may also be associated with anti-social behavior. This new variable is Drug Dependence in the past year. It is a 2 level categorical variable. After running the Chi Square test for each of the 3 cases with Drug dependence as a moderator, the following results have shown.

The above information tells that for those observations with alcohol dependence but no drug dependence, the association between alcohol dependence and anti-social behavior is highly significant. But for those observations with drug dependence, the association between alcohol dependence and anti-social behavior is weakened or insignificant. This is possible because drug dependence is another variable which can be highly associated with anti-social behavior. Therefore for observations with alcohol and drug dependence, the anti-social behavior can be attributed to both alcohol and drug dependence variables.

Other moderators like Major Depression, Level of Education or Working status did not have any effect on the association between alcohol dependence and anti-social behavior. 

  4. DISCUSSION

From the results produced in the study, the rate of alcohol dependence seems to be significantly increasing with the increase in level of consumption of alcohol. Moreover, the increase in the rate of alcohol dependence is getting higher in the consumption categories 3 and above. Therefore, answering the first question posed in the study, the approximate level of alcohol consumption that makes a person alcohol dependent is consumption of 1000 drinks or above per year, i.e. 2.73 or more drinks per day or 19.1 or more drinks per week.  The univariate analysis also shows that the average number of drinks consumed per year by the observations of the NESARC study was 50. This probably explains why most of the NESARC data (92.28%) were not alcohol dependent. However, this conclusion is made only on the basis of graphical analysis. As the data is observational, no causation can be drawn.

The association between alcohol dependence and anti-social behavior (3 types for this analysis) has been significant in all of the three cases. With majority of the alcohol dependents having a problem with their neighbors (70.48%); or their financial position (65.37%); or the police (78.63%), and this association being statistically significant in all 3 cases, a conclusion can be reached that there is a significant association between the alcohol dependence and the anti-social behavior in a person. For those observations with alcohol and drug dependence, the association is weakened or insignificant and the anti-social behavior can be attributed to both the variables. However, for the reason that the data are observational, no causation can be assigned.

4.1. Strengths

The NESARC sample is highly randomized and represents the population.

SAS Statistical Software has been used for analysis to minimize the arithmetic or statistical errors of manual analysis. 

4.2. Limitations

The data is observational and no causation can be made.

Only graphical analysis has been used to determine the approximate level of alcohol consumption that could trigger alcohol dependence.

4.3. Recommended Further Study

 The exact level of alcohol consumption leading to alcohol dependence using experimental data

The reason for alcohol dependents to behave anti-socially – Are there any other disorders/illnesses that arise from alcohol dependence that take charge of this behavior.

Further study on moderators like drug usage, medications, income levels or ethnicities that could change the association between alcohol dependence and anti-social behavior. 

5. REFERENCES

Passion Driven Statistics – Coursera.org

Wikipedia.com

National Institute on Alcohol Abuse and Alcoholism (www.niaaa.nih.gov)

 NESARC (http://aspe.hhs.gov/hsp/06/catalog-ai-an-na/nesarc.htm)

6. APPENDIX

6.1. SAS Program

  libname mydata "/courses/u_coursera.org1/i_1006328/c_5333" access=readonly;

Data new; set mydata.nesarc_pds;

  label

S2AQ8A="How often did you drink any alcohol in last 12 months (per year)"

USFREQYR = "How often did you drink any alcohol in last 12 months (per year)"

S2AQ8B="Number of drinks of any alcohol usually consumed on days when drank alcohol in last 12 months (per day)"

Numalcyr_est = "Estimated number of drinks consumed in total (per year) in the last 12 months"

S2AQ8C="Largest number of drinks of any alcohol consumed on days when drank alcohol in last 12 months (per day)"

ALCABDEP12DX="Alcohol abuse/dependence in last 12 months"

ALCABDEPP12DX="Alcohol abuse/dependence prior to the last 12 months"

S1Q239="Had a problem with neighbor, friend, or relative in last 12 months"

S1Q2310="Experienced a major financial crisis, bankruptcy, or been unable to pay bills on time in last 12 months"

S1Q2311="Did you or your family member have a trouble with police, got arrested, or sent to jail in last 12 months"

ConsumptionCategory="The category of consumption of alcohol per year in the last 12 months (0-5 ascending)"

AlcoholDependence="Alcohol abuse/dependence in last 12 months"

Social1="Had a problem with a neighbor, friend, or relative in last 12 months"

Social2="Experienced a major financial crisis, bankruptcy, or had been unable to pay bills on time in last 12 months"

Social3="Had a trouble with police, got arrested, or got sent to jail in last 12 months"

S1Q6A="Highest grade or year of School completed"

Edulevel="Highest grade or year of School completed"

S1Q7A1="Current working status"

Workstatus="Current working status"

MAJORDEPLIFE="Experienced major depression"

DrugUsage="Drug Dependence in the past 12 months"

;

  If S2AQ8A = 99 then S2AQ8A = . ;

If S2AQ8A = . then S2AQ8A = 11 ; /*Data management - setting unknown observations as missing and valid data under 11 variable. Variable 11 indicates those participants who are former drinkers or lifetime abstainers.*/

If S2AQ8B = 99 then S2AQ8B = . ;

If S2AQ8B = . then S2AQ8B = 0;/*Data management - setting unknown observations as missing and valid data under 0 variable. Variable 0 indicates those participants who are former drinkers or lifetime abstainers.*/

If S2AQ8C = 99 then S2AQ8C = . ;

If S2AQ8C = . then S2AQ8C = 0;/*Data management - setting unknown observations as missing and valid data under 0 variable. Variable 0 indicates those participants who are former drinkers or lifetime abstainers.*/

If S1Q239 = 9 then S1Q239 = . ;/*Data management - setting unknown observations as missing*/

If S1Q2310 = 9 then S1Q2310 = . ;/*Data management - setting unknown observations as missing*/

If S1Q2311 = 9 then S1Q2311 = . ;/*Data management - setting unknown observations as missing*/

  If S2AQ8A = 1 then USFREQYR = 365;

Else If S2AQ8A = 2 then USFREQYR = 264;

Else if S2AQ8A = 3 then USFREQYR = 168;

Else if S2AQ8A = 4 then USFREQYR = 120;

Else if S2AQ8A = 5 then USFREQYR = 60;

Else if S2AQ8A = 6 then USFREQYR = 30;

Else if S2AQ8A = 7 then USFREQYR = 12;

Else if S2AQ8A = 8 then USFREQYR = 9;

Else if S2AQ8A = 9 then USFREQYR = 4.5;

Else if S2AQ8A = 10 then USFREQYR = 1.5;

Else if S2AQ8A = 11 then USFREQYR = 0;

  Numalcyr_est = USFREQYR*S2AQ8B;/* numalcyr_est gives the total amount of alcohol drinks consumed in the past 12 months*/

/*To convert the quantitative variable into categories->*/

If Numalcyr_est eq 0 then ConsumptionCategory = 0;

Else If Numalcyr_est le 500 then ConsumptionCategory = 1;

Else if numalcyr_est le 1000 then ConsumptionCategory = 2;

Else if numalcyr_est le 5000 then ConsumptionCategory = 3;

Else if numalcyr_est le 10000 then ConsumptionCategory = 4;

Else if numalcyr_est gt 10000 then ConsumptionCategory = 5;

  /*To reduce the number of categories in a categorical variable->*/

If ALCABDEP12DX = 0 then AlcoholDependence = 0;/*0 is NO, there is no alcohol dependence*/

Else AlcoholDependence = 1;/* 1 is YES, there is alcohol dependence*/

  If S1Q239 = 1 then Social1 = 1;

Else if S1Q239 = 2 then Social1 = 0;

If S1Q2310 = 1 then Social2 = 1;

Else if S1Q2310 = 2 then Social2 = 0;

If S1Q2311 = 1 then Social3 = 1;

Else if S1Q2311 = 2 then Social3 = 0;

  /*categorising education levels*/

If S1Q6A le 7 then Edulevel = 1;/*Did not complete high school*/

Else If S1Q6A le 9 then Edulevel = 2;/*Completed high school or GED*/

Else If S1Q6A le 11 then Edulevel = 3;/*Some College or associate degree*/

Else If S1Q6A ge 12 then Edulevel = 4;/*Bachelor’s degree or higher*/

  /*categorising working status*/

If S1Q7A1 = 1 then Workstatus = 1;

Else If S1Q7A1 = 2 then Workstatus = 0;

  /*categorising drug usage*/

If OTHB12ABDEP = 0 then DrugUsage = 0;

Else DrugUsage = 1;

If HER12ABDEP = 0 then DrugUsage = 0;

Else DrugUsage = 1;

If MAR12ABDEP = 0 then DrugUsage = 0;

Else DrugUsage = 1;

If COC12ABDEP = 0 then DrugUsage = 0;

Else DrugUsage = 1;

  Proc sort; by IDNUM;

proc corr; var AGE numalcyr_est;

  Proc univariate; var USFREQYR numalcyr_est;

proc freq; tables USFREQYR numalcyr_est ConsumptionCategory AlcoholDependence Social1 Social2 Social3 DrugUsage/*S2AQ8A S2AQ8B S2AQ8C ALCABDEP12DX ALCABDEPP12DX S1Q239 S1Q2310 S1Q2311*/;

  /*Chi Square Tests*/

  proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

proc freq; tables Social1*AlcoholDependence/Chisq;

run;

proc freq; tables Social2*AlcoholDependence/Chisq;

run;

proc freq; tables Social3*AlcoholDependence/Chisq;

run;

  /*Moderation*/

  Proc sort; by DrugUsage;

proc freq; tables Social1*AlcoholDependence/Chisq; by DrugUsage;

run;

  Proc sort; by DrugUsage;

proc freq; tables Social2*AlcoholDependence/Chisq; by DrugUsage;

run;

  Proc sort; by DrugUsage;

proc freq; tables Social3*AlcoholDependence/Chisq; by DrugUsage;

run;

  Proc Anova;

Class DrugUsage;

Model numalcyr_est=DrugUsage;

Means DrugUsage;

  Proc sort; by Workstatus;

proc freq; tables Social1*AlcoholDependence/Chisq; by Workstatus;

run;

  Proc sort; by Workstatus;

proc freq; tables Social2*AlcoholDependence/Chisq; by Workstatus;

run;

  Proc sort; by Workstatus;

proc freq; tables Social3*AlcoholDependence/Chisq; by Workstatus;

run;

  Proc sort; by Edulevel;

proc freq; tables Social1*AlcoholDependence/Chisq; by Edulevel;

run;

  Proc sort; by Edulevel;

proc freq; tables Social2*AlcoholDependence/Chisq; by Edulevel;

run;

  Proc sort; by Edulevel;

proc freq; tables Social3*AlcoholDependence/Chisq; by Edulevel;

run;

  Proc sort; by MAJORDEPLIFE;

proc freq; tables Social1*AlcoholDependence/Chisq; by MAJORDEPLIFE;

run;

  Proc sort; by MAJORDEPLIFE;

proc freq; tables Social2*AlcoholDependence/Chisq; by MAJORDEPLIFE;

run;

  Proc sort; by MAJORDEPLIFE;

proc freq; tables Social3*AlcoholDependence/Chisq; by MAJORDEPLIFE;

run;

    /*Bonferroni adjustment*/

  Data Comparison1; set NEW;

If ConsumptionCategory = 0 or ConsumptionCategory=1;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison2; set NEW;

If ConsumptionCategory = 0 or ConsumptionCategory=2;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison3; set NEW;

If ConsumptionCategory = 0 or ConsumptionCategory=3;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison4; set NEW;

If ConsumptionCategory = 0 or ConsumptionCategory=4;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison5; set NEW;

If ConsumptionCategory = 0 or ConsumptionCategory=5;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison6; set NEW;

If ConsumptionCategory = 1 or ConsumptionCategory=2;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison7; set NEW;

If ConsumptionCategory = 1 or ConsumptionCategory=3;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison8; set NEW;

If ConsumptionCategory = 1 or ConsumptionCategory=4;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison9; set NEW;

If ConsumptionCategory = 1 or ConsumptionCategory=5;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison10; set NEW;

If ConsumptionCategory = 2 or ConsumptionCategory=3;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison11; set NEW;

If ConsumptionCategory = 2 or ConsumptionCategory=4;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison12; set NEW;

If ConsumptionCategory = 2 or ConsumptionCategory=5;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison13; set NEW;

If ConsumptionCategory = 3 or ConsumptionCategory=4;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison14; set NEW;

If ConsumptionCategory = 3 or ConsumptionCategory=5;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

run;

  Data Comparison15; set NEW;

If ConsumptionCategory = 4 or ConsumptionCategory=5;

Proc Sort; by IDNUM;

proc freq; tables AlcoholDependence*ConsumptionCategory/Chisq;

  run;

 6.2. Output