A Critical Analysis of Statistical Approaches in Journal Articles

The first paper was a journal article by Lam, Liang, Chikritzhs, & Allsop (2014) on “Alcohol and other drug use at school leavers' celebrations”, and the second one was the research on “The rising tide of diabetes mellitus in a Chinese population” by Wong, Leung, Tsang, and Griffiths (2013). The first paper was a journal article by Lam, Liang, Chikritzhs, & Allsop (2014) on “Alcohol and other drug use at school

A Critical Analysis of Statistical Approaches in Journal Articles

Abstract
Two journal articles have been reviewed by the scholar, and the results from
statistical analyses have been discussed. The first paper was a journal article by Lam, Liang,
Chikritzhs, & Allsop (2014) on “Alcohol and other drug use at school leavers' celebrations”,
and the second one was the research on “The rising tide of diabetes mellitus in a Chinese
population” by Wong, Leung, Tsang, and Griffiths (2013). The hypotheses of the studies,
sampling methodologies, and descriptive as well as inferential analyses have been discussed in
the present review article. A special attention has been set on the odd ratios (OR) of the
research papers, and an effort has been made to interpret the results based on OR. Finally,
limitations of both the articles have been discoursed.
2

Table of Contents
Abstract......................................................................................................................................2
A. Review of “Alcohol and other drug use at school leavers' celebrations” (Lam, Liang,
Chikritzhs, & Allsop, 2014).......................................................................................................4
1. Hypotheses and Variables...............................................................................................4
2. Sampling Methodology and its Advantages, Disadvantages..........................................5
3. Demographic Characteristics with Descriptive Statistics...............................................5
4. Inferential Statistics of the Study....................................................................................6
5. Explanation of Odds Ratio (OR).....................................................................................6
6. Arguments about Generalization of Results...................................................................7
B. Review of “The rising tide of diabetes mellitus in a Chinese population: A population-
based household survey on 121,895 persons” (Wong, Leung, Tsang, Griffiths, 2013)............7
1. Objectives and Variables.................................................................................................7
2. Demographic Characteristics from the Descriptive Statistics.........................................8
3. Elucidation on Inferential Statistics................................................................................8
4. Adjusted Prevalence Rates of Diabetes...........................................................................9
5. Odds Ratios for Self-Reported Diabetes.........................................................................9
6. Impact of the Limitations..............................................................................................10
References................................................................................................................................11
3

A. Review of “Alcohol and other drug use at school leavers' celebrations” (Lam,
Liang, Chikritzhs, & Allsop, 2014)
The primary objective of the studied article was to examine the influence of alcohol
and drug (AOD) consumption, focusing on the final year school students who were
attending the school leaving celebration. The focus was based on two objectives, the first
being the analysing the difference between AOD abuse in any social event and school
leaving celebration. Secondly, relation between harm-minimization policies and actual
consequences of AOD abuse was studied. Two set of hypotheses for the two objectives
were constructed with related variables.
1. Hypotheses and Variables
Objective A: Difference between AOD abuse in any social event and school leaving
celebration.
Null hypothesis: H0: Identical AOD abuse was present in students celebrating any
social event (peer-based) and school leaving celebration.
Alternate hypothesis: HA: Significantly dissimilar AOD abuse was found between
students celebrating any social event (peer-based) and school leaving celebration.
Independent Variable: Celebration type (any social event, school leavers’
celebration).
Dependent Variable: AOD usage
Objective B: Relation between harm-minimization policies and actual consequences
of AOD abuse was studied.
Null hypothesis: H0: There existed no relation between harm-minimization policies
and actual consequences of AOD abuse.
Alternate hypothesis: HA: Significantly relation existed between harm-minimization
policies and actual consequences of AOD abuse.
Independent Variable: Assignation in harm-minimization policies
4

Dependent Variable: Expected consequences from AOD abuse
2. Sampling Methodology and its Advantages, Disadvantages
The sampling methodology used in the reviewed research was a two-fold design
survey with convenience selection strategy. The participating students in the survey were
asked to report about opinion about AOD abuse and impact of harm-minimization
strategies in minimization of substance abuse. A psychometrically validated
questionnaire was used to collect data in face to face survey as well as through online
surveys. As, comparison between AOD abuse was the objective of the study, sample data
was collected from ordinary social event (peer-based) and school leavers’ celebration.
Among advantages of convenience selection strategy was cost efficiency and large
sample collection. Self-reported design of the survey ensured that data was collected
from first hand source, and the large sample ensured the validity of the data.
Among the disadvantages, reliability of the self-reported data about AOD abuse was a
major issue. Students might have provided answers carefully to avoid any tarnishing
impact whatsoever the feared. Also, overstated answers were also possible due to some
over policed students.
3. Demographic Characteristics with Descriptive Statistics
Statistics about gender, and age were available in the studied research. For pairwise
comparison purpose, sample data was collected from 541 (37% of total population)
students (female = 56%) intended to attend school leavers’ celebration, and also from
405 (28% of total population) students (female = 50%) post their celebration. In 52%
cases pre-celebration (last social event) data was collected from online responses,
whereas in 86% cases face to face sample data collection was done post-celebration
(school leavers’ celebration), primarily to reduce any possible error occurring from
unintentional incorrect information due to forgetting because of gap of time. Pre-
celebration sample contained 91% students aged 17 years, and the rest 9% students were
18 years of age. Post-celebration sample consisted of 94% 17 years students, with rest
being just 18 years old. Paired comparison was possible with data found for 120
participants (female = 62%).
5

4. Inferential Statistics of the Study
Objective A, difference between AOD abuse in any social event and school leaving
celebration was tested with numerous “Wilcoxon signed-rank test” as an alternative to
paired t-test, due to possibility of skewed data from self-reported data collection.
“Wilcoxon signed-rank test” is a non-parametric test which is used for pairwise
comparison of non-normal data. Also, the sample data collection was done using a Likert
scale questionnaire, implying the data collected was ordinal in nature. Hence, choice of
inferential test was appropriate and justified.
Objective B, relation between harm-minimization policies and actual consequences
of AOD abuse was studied using Logistic regression models. The impact of certain
variables, such as AOD use, safety policies, gender of participants, and location of stay
on expected consequences from AOD use was measured.
Logistic regression models are generally used where the dependent variable is
categorical in nature. The linear regression model fails to assess the impact of predictors
on categorical outcome variable, especially the individual impact on all the categories.
The odd ratios in the Logistic regression models display the exact relation with the
predictors, especially with reference to referent category.
5. Explanation of Odds Ratio (OR)
Odd ratios (OR) in Logistic regression models reflect the constant effect of the
independent variable on the likelihood of occurrence of the outcome variable. OR of
participants with lowest frequency of safety strategy for engaging unprotected sex was
10.92 with respect to the participants with highest frequency of safety strategies from
AOD abuse. Interestingly, the result did not include the impact of quantity of alcohol/
drug, gender, location, and survey type (controlled factors). Hence, students who adopted
lowest safety measures were at 10.92 or roughly 11 times higher risk of suffering from
adverse impact of AOD use, even without intervention of other factors.
6

6. Arguments about Generalization of Results
The sample data collected from students attending celebration parties in Rottnest
Island, located at 20 km off the west coast of Perth, Australia. The data was primarily
collected from students who were just old (17 years) to purchase alcohol legally. Overall,
almost over 90% students were of 17 years of age with rest being 18 years. The sample
size was almost 1/3rd of students’ population in Rottnest Island. Hence, generalization of
results for national population of school goers was possible. But, two factors were
identified from the study, which were potential obstructions behind generalization of
results. Firstly, survey on school leavers makes the sample age specific which reduces
the heterogeneity and validity of results. Secondly, study with self-reported sample data
always has a possibility of being highly skewed, and thus the results of this particular
study could be year (2009) and area (Western Australia) specific.
B. Review of “The rising tide of diabetes mellitus in a Chinese population: A
population-based household survey on 121,895 persons” (Wong, Leung,
Tsang, Griffiths, 2013).
Primary objective of the studied paper was to assess the generality of results found
from analysing the effect of age, household income, and sex on diabetes prevalence
among 121,895 participants representing entire Hong Kong population (Census and
Statistics Department, 2010). The survey was conducted in 2001, 2002, 2005, and 2008
to evaluate results across a period of 8 years. The entire sample was stratified in two
strata based on gender of the participants. A binary logistic regression model was used
for the purpose.
1. Objectives and Variables
Objective A was to assess the effect of increase in age on diabetes prevalence among
the participants.
Null hypothesis: H0: There existed no association between increase in age and
diabetes prevalence.
Alternate hypothesis: H0: There existed statistically significant association between
increase in age and diabetes prevalence (0-39 was referent age group).
7

Objective B was to assess the effect of low household income on diabetes prevalence
among the participants.
Null hypothesis: H0: There existed no association between low household income
and diabetes prevalence.
Alternate hypothesis: H0: There existed statistically significant association between
low household income and diabetes prevalence (participants earning above $ 50,000
referent income group).
2. Demographic Characteristics from the Descriptive Statistics
Diabetes prevalence of 121,895 people across 2001, 2002, 2005, and 2008 was
collected with demographic information regarding their age, household income, and
gender. The sample consisted of 103,367 adult participants with age of 15 years and
more. The average age of participants in the sample was calculated to be 38.2 years.
Information on gender of 121,895 participants revealed a balances presence of both
the genders with females (N = 61, 831, 50.2%) being just greater in number. Household
income (HK dollars) of sample participants was categorised in four categories (≥ 50,000,
25,000-49,999, 10,000-24,999, and ≤ 9,999). Presence of 10,000-24,999 income group of
participants was the highest (N = 50,648, 42.4%), followed by 10,000-24,999 income
group (N = 32,748, 27.4%), ≤ 9,999 (N = 23,578, 19.7%), and ≥ 50,000 (N = 12,452,
10.4%).
Sample was categorized according to age (years) in eight groups (< 15, 15-24, 25-
34, 35-44, 45-54, 55-64, 65-74, and ≥ 75). Among 103,367 adult participants (≥ 15),
13.8% (N = 16, 834) belonged to age group of 15-24, 14.6% (N = 17,751) to age group
of 25-34, 18.2% (N = 22,206) to age group of 35-44, 16.4% (N = 20,033) to age group
of 15-24, 9.2% (N = 11,179) to age group of 15-24, and a total of 12.6% (N = 15,364)
belonged to age groups of 65-74, and ≥ 75.
3. Elucidation on Inferential Statistics
Inferential analysis for evaluating the impact of age and income on diabetes
prevalence across years was Binary Logistic Regression. In the constructed model age
8

and groups were adjusted for better comparison. The age group of 0-39 years was the
referent, whereas income group of ‘≥ 50,000’ was considered as the referent in the
regression model. A multivariate regression model was also used to assess the
independent association between diabetes prevalence and participants’ demographic
details.
Initially, use of multivariate regression model indicated the causal relation and
association between diabetes and demographic factors. Binary Logistic Regression
models are generally used where the dependent variable has two categories. The linear
regression model fails to assess the impact of predictors on two different categories of an
outcome variable. The odd ratios in the Binary Logistic regression models display the
exact relation with the predictor, especially with reference to age and income referent
categories (Harrell, 2015).
4. Adjusted Prevalence Rates of Diabetes
The results in the Binary Logistic Regression model were statistically significant
when age and sex were adjusted for measuring diabetes prevalence. Two separate
regression models were constructed based on gender, and in each model age groups were
reorganized to better comparison of diabetes prevalence. Importantly, the study also
considered 2001 as base year or referent year to compare the results of 2005 and 2008.
Initially, females were noted to be (31.8%, 2005; 69.3%, 2008) have higher diabetes
prevalence compared to that of the males (27.8%, 2005; 47.9%, 2008). But, when
adjusted for sex no significant difference in diabetes prevalence was noted between male
and females. Also, significantly increasing diabetes prevalence was noted for lower
household income group when compared to highest income group.
5. Odds Ratios for Self-Reported Diabetes
Adjusted Odd Ratio (AOR) for sex and age were evaluated from the Logistic
Regression Model. Age adjusted groups comparison revealed that people aged between
40 and 65 years (AOR = 32.21, 95% CI 20.6–50.4, p < 0.001) were significantly at
higher risk of diabetes prevalence compared to the referent age group of 0-39 years.
9

Notably, people aged over 65 years were 120 times more associated (AOR = 120.1, 95%
CI 76.6–188.3, p < 0.001) to diabetes compared to referent group.
Monthly household income category of 25,000-49,999 (AOR = 1.39, 95% CI 1.04-
1.86, p < 0.05), 10,000-24,999 (AOR = 1.58, 95% CI 1.2-2.07, p < 0.001), and ≤ 9,999
(AOR = 2.19, 95% CI 1.66-2.88, p < 0.001) were all significantly at a higher risk of
association with diabetes compared to highest income group (≥ 50,000), especially the
lowest income group had almost two-fold chance of diabetes in such comparison.
6. Impact of the Limitations
The coefficient of determination in the Binary Logistic Regression model was R2 =
0.198, implying that adjusted variables were able to explain 19.8% variation in diabetes
prevalence. Hence, search of other predictors of diabetes prevalence, such as eating
habit, family history, and affinity towards sugar and carb would have been beneficial.
Also, it has to be noted that the sample data was collected from self-reported survey of
Chinese people. From previous literatures, it can be illustrated that most of the people in
China are ignorant about preventive diabetes check-up (Yang et. al., 2010). Therefore,
the self-reported data could have been erroneous and skewed. Generalization of the
statistical analyses of the study could be a terrible mistake.
10

References
Census and Statistics Department (2010) The Government of the Hong Kong special
administrative region. http://www.censtatd.gov.hk/hong_kong_statistics/social_
topics_ studies/index.jsp.Accessed30 Aug 2010.
Harrell, F. E. (2015). Ordinal logistic regression. In Regression modeling strategies (pp. 311-
325). Springer, Cham.
Lam, T., Liang, W., Chikritzhs, T., & Allsop, S. (2014). Alcohol and other drug use at school
leavers' celebrations. Journal of Public Health, 36(3), 408-416.
Wong, M. C., Leung, M. C., Tsang, C. S., Lo, S. V., & Griffiths, S. M. (2013). The rising tide
of diabetes mellitus in a Chinese population: a population-based household survey on
121,895 persons. International journal of public health, 58(2), 269-276.
Yang, W., Lu, J., Weng, J., Jia, W., Ji, L., Xiao, J., ... & Zhu, D. (2010). Prevalence of
diabetes among men and women in China. New England Journal of
Medicine, 362(12), 1090-1101.
11