Data Science

In the context of the regression modeling practice course from Wesleyan University, I am going to test a basic linear regression model. The informations about the sample, procedure and measures used in this post are detailed in the previous post (here).

The response variable (alcohol abuse/dependence) is caterogical with 4 scales (0 to 3). For this assignment, the variable has been considered as quantitative. The explanatory variable used in this assignment is the major depression, which is a categorical variable with 2 categories (0=No and 1=Yes).

Program

The image below shows the  program I used to subset my data (employed individuals aged from 30 to 40), create some variables to better calculate what I want to analyze (alcohol abuse/dependence during lifetime), and convert the categorical scales of my response variable into quantitative variable.

After the completion of these data, I created a frequency table for the chosen explanatory variable (major depression) and tested a linear regression.

Results output

The frequency table shows that there are more individuals who did not face major depression during lifetime (variable value = 0).

The results of the linear regression model indicates that major depression has a positive correlation with alcohol abuse/dependence (beta = 0.5 and p=3e-68).

Sample

I want to focus my study on the relationship between alcohol dependence/abuse and major depression within employed population. For this purpose, the sample has been taken from the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC). The survey was conducted on 43,093 participants that covered drug and alcohol disorders prevalence, risk factors, environment interaction disorders and disabilities and psychiatric disorders.

The data analytic sample for this study includes individuals aged between 30 and 40 years old who are currently employed (N=7’996) – 87.3% are employed full-time and 12.7% part-time.

Procedure

The data was collected by U.S. Census Bureau Field Representatives during 2001– 2002 all around the USA. The information has been collected through computer-assisted personal interviews (CAPI) in the participant’s home. This survey was originally conducted to evaluate the level and have a better understanding of the association between alcohol use and other drug use and/or psychiatric disorders. This knowledge could help to find more effective treatments for psychiatric patients.

Measures

The assessment of major depression has been made using the DSM-IV Diagnoses: major depression period(s) in lifetime or in the last 12 months. This response variable is considered as categorical (yes/no).

Alcohol dependence or abuse was evaluated in the last 12 months at the time of the interview or prior to the last 12 months. These datas are alle available in the DSM-IV Diagnoses survey part.

In order to determine if the parental alcoholic behaviour background moderate the number of years of dependence onset on individuals starting drinking before 21 years old, a chi-square test has been conducted.

Here is the code performing the testing:

Chi-Square Test :

When examining the association between the age at which the individuals started drinking (grouped age - categorical response) and the number of years of alcohol dependence onset (grouped number of years - categorical explanatory), the average number of years of dependence onset is around 12 years for the individuals who started drinking before 12 years old and around 7 years for those who started drinking between 12 and 21 years old.

The Chi Square test of independence revealed that among individuals who started drinking alcohol before 21 years old, the starting age of drinking and the number of years of dependence onset are significantly associated (X2 =182.60, p=2.93e-35).

When considering the parental alcoholic behaviour (moderator):

For individuals who do not have alcoholic parents: the Chi Square test of independence revealed that the starting age of drinking and the number of years of dependence onset are significantly associated (X2 =83.00, p=1.21e-14).

For individuals who have alcoholic parents: the Chi Square test of independence revealed that the starting age of drinking and the number of years of dependence onset are significantly associated (X2 =85.75, p=3.39e-15).

In order to determine if the the age at which the individuals started drinking has an impact on the number of years of alcohol dependence onset, a pearson correlation has been done.

Here is the code performing the testing:

Below are the scatterplot associated to the script:

Below is the output of the code on IPython:

Model Interpretation for Pearson correlation test:

When examining the association between the age of alcohol dependence onset of the individuals  (age - quantitative response variable) and the age at which the individuals started drinking (age – quantitative explanatory variable), a Pearson correlation testing revealed that the relationship is statistically not significant: p > 0.05. The Pearson correlation coefficient shows a weak positive relationship between the variables.

When examining the association between the number of years of alcohol dependence onset after first drinks  (number of years - quantitative response variable) and the age at which the individuals started drinking (age – quantitative explanatory variable), a Pearson correlation testing revealed that the relationship is statistically not significant: p > 0.05. The Pearson correlation coefficient shows a weak negative relationship between the variables.

In order to determine if the parental background of individuals starting drinking before 21 years old has an impact on the categories of age at which they started, a chi-square analysis has been done.

Here is the code performing the testing:

Below is the output of thwe code on IPython:

Model Interpretation for Chi-Square Tests:

When examining the association between the age at which the individuals started drinking (grouped age - categorical response) and the parental background (alcoholic or not - categorical explanatory), a chi-square test of independence revealed that among individuals who have alcoholic parents, 35.9% started drinking before age 15, compared to those who does not have alcoholic parents (24.4%), X2 =67.99, 1 df, p=1.64e-16.

The df or degree of freedom we record is the number of levels of the explanatory variable -1. Here the df is 1 alcoholic parents which has 2 levels (df 2-1=1).

To determine the association between the variables of interest, an hypothesis testing have been run using ANOVA and Tukey’s Honestly Significant Difference test on Python:

ANOVA testing to determine the association between having one or both biological alcoholic parents and the number of years of the alcohol dependence onset.

Null hypothesis (Ho): there is no relationship between having one or both biological alcoholic parents and the number of years of alcohol dependence onset after the first drink.

Alternative hypothesis (Ha): having one or both biological alcoholic parents and the number of years of alcohol dependence onset after the first drink are related

The F value is at 2.151 and the p at 0.116. As the p-value is higher than 0.05, we can conclude the data doesn’t provide enough evidence that there is a correlation between the considered variables. These values show that having one or both biological alcoholic parents has no influence in the number of years of alcohol dependence onset after first drink.

ANOVA testing to determine the association between having one or both biological alcoholic parents  and the age at which individuals start drinking

Null hypothesis (Ho): there is no relationship between having one or both biological alcoholic parents  and the age at which the individuals start

Alternative hypothesis (Ha): having one or both biological alcoholic parents  and the age at which the individuals start drinking alcohol are related

In order to understand the variables taken into account for my study, univariate graphs have been made with the following code. Other graphs have been also made to see the association of the variables between them.

 Here is the univariate graph representing the age at which the individuals started drinking alcohol:

This graph can be considered as bimodal, with the highest peak at 18 years old and the second highest peak at 21 years old. Anyway, the graph is skewed to the left as there are higher frequency in higher categories, meaning that the individuals tend to start drinking at higher age.

Here is the univariate graph representing the number of years of alcohol dependence onset after the individuals started drinking:

This graph is unimodal and skewed at the right as there are higher frequency in lower number of years of alcohol dependence onset. The highest peak is at 0 year (the individuals counted in this peak developed alcohol dependence in less than 1 year after they started drinking).

Here is the univariate graph representing the individuals having one or both of their blood parents with alcohol problems:

 The graph is unimodal with its highest peak for “No”, which means the individuals of whom the parents are not alcoholics or problem drinkers.

 The next graph represents the association between the age at which the individuals started drinking alcohol and the number of years of their alcohol dependence onset.

The aim of the study is to know how fast the alcohol dependence is set on when starting drinking before age of 21. To this purpose, I created a new variable called ONSETASTART which is the number of years between the start drinking year and the onset of alcohol dependence year. To be easier to read, the datas have been subsetted in different starting age class (0-5, 5-10, 10-15, and so on) and in different class of the years of onset. I also subtracted all the life abstainer induciduals from the dataset.

Here is the code for the data management:

And here is the output on IPython console:

43’093 individuals have been asked at which age they started drinking, not counting small tastes or sips. Of the total individuals, about 63.8% started drinking before or at age 21, which is the legal age of starting strong alcohol. When asked the age at onset of alcohol dependence, of the total individuals, around 6.1% started being dependent before or at age 21.

The individuals who started drinking before or at age 21 were asked if their father was ever an alcoholic or problem drinker. Around 22.0% answered yes and 72.7% answered no. The remaining 5.3% are unknown.

I focused my research on the U.S. National Epidemiological Survey on Alcohol and Related Conditions (NESARC) -  a survey designed to determine the magnitude of alcohol use and psychiatric disorders in the U.S. population.

After looking at the codebook of the NESARC study, I decided to focus on the alcoholism part. The subject is so vast that I have included in my first codebook the whole section dealing with alcoholism. We all know that alcohol dependence is problematic if no treatment or prevention is put in place. This is problematic for the physical health of the individual but also for his psychological health, his relationships and his finances.

In terms of prevention, I wondered if an individual who begins to drink early in age develops more quickly alcohol dependence. Some people will quickly become dependent while others manage to do without it anyway. 

So I will try to make the connection between the age at which the individual begins to drink and the age at which he develops dependence: how fast does it occur?

The second question that arose in association of this topic is the genetic influence in the tendency to quickly develop alcohol dependence.

To explore all these topics, I added to my codebook variables about the ages of starting drinking and starting being dependent as well as variables reflecting the blood family history of alcoholism.

According to the study made by Ralph W. Hingson (1), individuals beginning to drink before age 14 tend to become dependent within 10 years after first drinking. I have not found any other research on the time between drinking start and dependence onset. Other articles (2) clearly show the effect of family history of alcoholism and the development of alcohol dependence, but not the rate at which the individual develops a dependence.

(1)  Ralph W. Hingson et al,  Age at Drinking Onset and Alcohol DependenceAge at Onset, Duration, and Severity,  Arch Pediatr Adolesc Med. 2006;160(7):739-746