Yours Sincerely @giuliojcg - Tumblr Blog

Testing a Potential Moderator

I analysed the relationship between female employee rate and freedom (democracy score), with income level acting as a moderator variable. It is expected that as the democracy index is higher, the level of female employment shall rise too.

In the SAS output below we can see clearly that for low income categories (1 and 2) correlations are not significant, yet for higher income countries a strong relation is verified between female employee rate and democracy index.

SAS Code used

LIBNAME mydata "/courses/d1406ae5ba27fe300 " access=readonly; Data Analysis2; Set mydata.gapminder;

Label incomeperperson='Income per Capita' HIVrate='Estimated HIV Prevalence %' Internetuserate='Internet users (per 100 people)' breastcancerper100TH='Breast cancer new cases per 100,000 female' lifeexpectancy='Life expectancy at birth (years)' polityscore='Democracy score (Polity)' Income_Level='Income per Capita by Category' suicideper100TH='2005 Suicide, age adjusted, per 100 000' armedforcesrate='Armed forces personnel (% of total labor force)' Income_Level='Income per Capita by Category' breastcancerper100th_G= 'Breast cancer new cases per 100,000 female by Group' lifeexpectancy_G= 'Life expectancy interval' polityscore_G='Polity Score by Group' internetuserate_G='Internet users (per 100 people)by Group'; /*Data Management for incomeperperson*/ If incomeperperson <= 700 then Income_Level =1; else if incomeperperson > 700 and incomeperperson <= 5000 then Income_Level=2; else if incomeperperson > 5000 and incomeperperson<=10000 then Income_Level =3; else if incomeperperson > 10000 then income_level=4; else Income_Level=.; /*Data Management for lifeexpectancy*/ if lifeexpectancy >= 0 and lifeexpectancy <= 68 then lifeexpectancy_G = " 0-68" ; else if lifeexpectancy > 68 and lifeexpectancy <= 74 then lifeexpectancy_G ="68-74" ; else if lifeexpectancy > 74 then lifeexpectancy_G = "> 76" ; else if lifeexpectancy = . then lifeexpectancy_G = "NA" ; /*Data Management for polityscore*/ if polityscore >= -10 and polityscore <= -4 then polityscore_G = "-10 and -4" ; else if polityscore > -4 and polityscore <= 4 then polityscore_G ="-4 and 4" ; else if polityscore > 4 and polityscore <= 7 then polityscore_G ="4 and 7" ; else if polityscore > 7 and polityscore <= 9 then polityscore_G ="7 and 9" ; else if polityscore > 9 then polityscore_G = "> 9" ; else if polityscore = . then polityscore_G = "NA"; /*Data Management for internetuserrate*/ if internetuserate >= 0 and internetuserate <= 7.5 then internetuserate_G = "0 - 7.5" ; else if internetuserate > 7.5 and internetuserate <= 20 then internetuserate_G ="7.5-20"; else if internetuserate > 20 and internetuserate <= 41 then internetuserate_G = "20-41" ; else if internetuserate > 41 and internetuserate <= 66 then internetuserate_G = "41-66" ; else if internetuserate > 66 then internetuserate_G = ">66" ; else if internetuserate = . then internetuserate_G = "NA"; /*Data Management for breastcancerper100th*/ if breastcancerper100th >= 0 and breastcancerper100th <= 19 then breastcancerper100th_G = " 0 and 19" ; else if breastcancerper100th > 19 and breastcancerper100th <= 26 then breastcancerper100th_G = "19 and 26" ; else if breastcancerper100th > 26 and breastcancerper100th <= 34 then breastcancerper100th_G = "26 and 34" ; else if breastcancerper100th > 34 and breastcancerper100th <= 53 then breastcancerper100th_G = "34 and 53" ; else if breastcancerper100th > 53 then breastcancerper100th_G = ">53" ; else if breastcancerper100th = . then breastcancerper100th_G = "NA"; /*Urban Rate Groups*/ If urbanrate >= 50 then urban=1; else if urbanrate <50 then urban=0; else urban=99;

Proc Sort Data=Analysis2; by income_level;Run;

/*proc freq data=analysis2;tables income_level;run;*/

PROC CORR; VAR femaleemployrate polityscore; BY income_level;Run;

Generating a Correlation Coefficient in SAS

Scatter Diagram

The variables of interest in this assignment are Income per Capita, Internet users (per 100 people), Breast cancer new cases per 100,000 female, Life expectancy at birth (years) and Armed forces personnel (% of total labor force).

A preliminary analysis is done by plotting a scatter diagram for selected variables.

In this graph it is possible to infer a positive (linear?) relationship between life expectancy and internet users rate. In the next graph it is shown a positive and somewhat strong relationship between breast cancer new cases and internet users rate. This is a quite intriguing relationship. It is going to be further discussion about this in the next section when the correlation matrix is presented.

Coefficient of Correlation

The scatter diagram graphically described the relationship between some of the variable of interests. To do the same, but numerically we produced the coefficient of correlation matrix shown below.

From the correlation coefficient between life expectancy and internet user rate we can confirm that there is a strong and positive relationship between these two variables. (r=0.76392 with a p-value < 0.001). However, we cannot assume that there might be a casual relationship. However, it is possible to say that with the internet user rate we can explain on average almost 60% of life expectancy at birth (r2=0.583573766). Another interesting and apparently counterintuitive fact in this data set is the sign of the relationship between Internet users (per 100 people) and Breast cancer new cases per 100,000 female. As it was already pointed out in the scatter diagram section of analysis, there seems to be a positive (and strong relationship) between the mentioned variables!. The correlation coefficient confirm this with a r=0.78842 and a p<0.001. So, we can say that around 60% of breast cancer new cases is explained by internet user rate? However, we have to take carefully, because there might be in the presence of a confounding bias. More data analysis is needed to understand this phenomena. Perhaps, women with more access to internet (and information in general) are more willing to perform early diagnosis test, thus more women are appear to be as a new case. On the contrary in countries with less access to internet (and less income and education surely) the number of women that perform the diagnosis are inferior and therefore appear as “missing” data.

Appendix SAS Programs used

LIBNAME mydata "/courses/d1406ae5ba27fe300 " access=readonly; Data Analysis2; Set mydata.gapminder; Label incomeperperson='Income per Capita' HIVrate='Estimated HIV Prevalence %' Internetuserate='Internet users (per 100 people)' breastcancerper100TH='Breast cancer new cases per 100,000 female' lifeexpectancy='Life expectancy at birth (years)' polityscore='Democracy score (Polity)' Income_Level='Income per Capita by Category' suicideper100TH='2005 Suicide, age adjusted, per 100 000' armedforcesrate='Armed forces personnel (% of total labor force)' Proc Sort;by country;Run;

Proc GPLOT Data=Analysis2; Plot lifeexpectancy*Internetuserate; Proc GPLOT Data=Analysis2; Plot breastcancerper100th*Internetuserate; Proc GPLOT Data=Analysis2; Plot lifeexpectancy*polityscore; Proc GPLOT Data=Analysis2; Plot Internetuserate*polityscore; Proc GPLOT Data=Analysis2; Plot polityscore*Internetuserate; Proc GPLOT Data=Analysis2; Plot polityscore*incomeperperson; Proc GPLOT Data=Analysis2; Plot suicideper100TH*polityscore; Proc GPLOT Data=Analysis2; Plot suicideper100TH*incomeperperson; Proc GPLOT Data=Analysis2; Plot HIVrate*Internetuserate; Proc GPLOT Data=Analysis2; Plot lifeexpectancy*armedforcesrate;

PROC CORR Data=Analysis2; VAR incomeperperson lifeexpectancy Internetuserate breastcancerper100th polityscore suicideper100TH HIVrate armedforcesrate;run;

Running a Chi-Square Test of Independence

Model Interpretation for Chi-Square Tests

When analysing the association between degree of urbanisation in a country (categorical response) and income level (categorical explanatory), there is sufficient evidence to infer that the proportions of people living in urban areas with income level above median (67.21%) differ significantly from those with income level below the median average (32.79%).

Model Interpretation for post hoc Chi-Square Test results

There is strong evidence to conclude that urbanisation depends strongly on income level (collapsed into 3 groups).

On the other hand, using Post Hoc comparisons for each pair of income level (3 categories) and degree of urbanisation (2 categories), we found sufficient evidence to infer that the level of urbanisation increases as income levels rise up for all income categories (Low, Middle and High income countries).

LIBNAME mydata "/courses/d1406ae5ba27fe300 " access=readonly; Data Analysis2; Set mydata.gapminder; Length Income_Level2 $13;

Label incomeperperson='Income per Capita' HIVrate='Estimated HIV Prevalence %' Internetuserate='Internet users (per 100 people)' breastcancerper100TH='Breast cancer new cases per 100,000 female' lifeexpectancy='Life expectancy at birth (years)' polityscore='Democracy score (Polity)' Income_Level='Income per Capita by Category' breastcancerper100th_G= 'Breast cancer new cases per 100,000 female by Group' lifeexpectancy_G= 'Life expectancy interval' polityscore_G='Polity Score by Group' internetuserate_G='Internet users (per 100 people)by Group'; /*Data Management for incomeperperson*/ if incomeperperson >= 0 and incomeperperson <= 560 then Income_Level = "US$ 0-560 "; else if incomeperperson > 560 and incomeperperson <= 1850 then Income_Level ="US$ 560-1,850"; else if incomeperperson > 1850 and incomeperperson <= 5000 then Income_Level = "US$ 1,850-5,000"; else if incomeperperson > 5000 and incomeperperson <= 14000 then Income_Level = "US$ 5,000-14,000"; else if incomeperperson > 14000 and incomeperperson then Income_Level = ">US$ 14,000"; else Income_Level='NA'; /*Data Management for lifeexpectancy*/ if lifeexpectancy >= 0 and lifeexpectancy <= 68 then lifeexpectancy_G = " 0-68" ; else if lifeexpectancy > 68 and lifeexpectancy <= 74 then lifeexpectancy_G ="68-74" ; else if lifeexpectancy > 74 then lifeexpectancy_G = "> 76" ; else if lifeexpectancy = . then lifeexpectancy_G = "NA" ; /*Data Management for polityscore*/ if polityscore >= -10 and polityscore <= -4 then polityscore_G = "-10 and -4" ; else if polityscore > -4 and polityscore <= 4 then polityscore_G ="-4 and 4" ; else if polityscore > 4 and polityscore <= 7 then polityscore_G ="4 and 7" ; else if polityscore > 7 and polityscore <= 9 then polityscore_G ="7 and 9" ; else if polityscore > 9 then polityscore_G = "> 9" ; else if polityscore = . then polityscore_G = "NA"; /*Data Management for internetuserrate*/ if internetuserate >= 0 and internetuserate <= 7.5 then internetuserate_G = "0 - 7.5" ; else if internetuserate > 7.5 and internetuserate <= 20 then internetuserate_G ="7.5-20"; else if internetuserate > 20 and internetuserate <= 41 then internetuserate_G = "20-41" ; else if internetuserate > 41 and internetuserate <= 66 then internetuserate_G = "41-66" ; else if internetuserate > 66 then internetuserate_G = ">66" ; else if internetuserate = . then internetuserate_G = "NA"; /*Data Management for breastcancerper100th*/ if breastcancerper100th >= 0 and breastcancerper100th <= 19 then breastcancerper100th_G = " 0 and 19" ; else if breastcancerper100th > 19 and breastcancerper100th <= 26 then breastcancerper100th_G = "19 and 26" ; else if breastcancerper100th > 26 and breastcancerper100th <= 34 then breastcancerper100th_G = "26 and 34" ; else if breastcancerper100th > 34 and breastcancerper100th <= 53 then breastcancerper100th_G = "34 and 53" ; else if breastcancerper100th > 53 then breastcancerper100th_G = ">53" ; else if breastcancerper100th = . then breastcancerper100th_G = "NA"; /*Urban Rate Groups*/ If urbanrate >= 50 then urban=1; else if urbanrate <50 then urban=0; else urban=99;

/*Income per person median*/ If incomeperperson >= 2549.5584738 then Median_income='Above Median Income'; else if incomeperperson <2549.5584738 then Median_income='Below Median Income'; else Median_Income='No Data';

/*Income per person The Middle*/ If incomeperperson <= 1295 then Income_Level2='Low Income'; else if incomeperperson > 1295 and incomeperperson <= 5900 then Income_Level2='Middle Income'; else if incomeperperson > 5900 then Income_Level2='High Income'; else Income_Level2='No Data';

keep country incomeperperson breastcancerper100TH HIVrate Internetuserate lifeexpectancy polityscore urbanrate Income_Level lifeexpectancy_G polityscore_G internetuserate_G breastcancerper100th_G urban Median_Income Income_Level2;

Proc Sort;by country;Run; Proc Freq Data=analysis2;Tables Income_Level2;Run; Proc Freq Data=analysis2;Tables urban*Median_income/ChisQ;Run;

Proc Freq Data=analysis2;Tables urban*Income_Level2/ChisQ;Run;

Data COMPARISON1; Set Analysis2; If Income_Level2='Low Income' or Income_Level2='Middle Income'; Proc Sort; by country; Proc Freq; Tables urban*Income_Level2/ChisQ; Run; Data COMPARISON2; Set Analysis2; If Income_Level2='Low Income' or Income_Level2='High Income'; Proc Sort; by country; Proc Freq; Tables urban*Income_Level2/ChisQ; Run; Data COMPARISON3; Set Analysis2; If Income_Level2='Middle Income' or Income_Level2='High Income'; Proc Sort; by country; Proc Freq; Tables urban*Income_Level2/ChisQ; Run;

#chisquare #statistics #SAS

Running an analysis of variance in SAS

Two relationships are tested using ANOVA. In the first place, it is analysed the internet use rate (quantitative response) for countries with the majority of its population living in urban (categorical explanatory variable) areas compared with those with less than 50% of people living in urban areas. The variance analysis shows that there are significative difference in the internet use rate between non urban countries (Mean=18.89, s.d. ±21.51) and urban countries (Mean=47.08, s.d. ±25.75), F(1, 190)=63.27, p<0001.

The second part of the analysis look for relationships between access to internet (quantitative variable) and the democratic and free nature of countries (explanatory variable collapsed into 4 groups, -10 is the lowest value, 10 the highest.). The Post Hoc analysis shows that the internet use rate in countries with more than a score of 7 have a mean significantly different (greater) than countries with scores between -4 and 4. Note however that means for countries with score between 7-9 and -10 and -4 are not significantly different.

Code:

LIBNAME mydata "/courses/d1406ae5ba27fe300 " access=readonly; Data Analysis2; Set mydata.gapminder;

Label incomeperperson='Income per Capita' HIVrate='Estimated HIV Prevalence %' Internetuserate='Internet users (per 100 people)' breastcancerper100TH='Breast cancer new cases per 100,000 female' lifeexpectancy='Life expectancy at birth (years)' polityscore='Democracy score (Polity)' Income_Level='Income per Capita by Category' breastcancerper100th_G= 'Breast cancer new cases per 100,000 female by Group' lifeexpectancy_G= 'Life expectancy interval' polityscore_G='Polity Score by Group' internetuserate_G='Internet users (per 100 people)by Group'; /*Data Management for incomeperperson*/ if incomeperperson >= 0 and incomeperperson <= 560 then Income_Level = "US$ 0-560 "; else if incomeperperson > 560 and incomeperperson <= 1850 then Income_Level ="US$ 560-1,850"; else if incomeperperson > 1850 and incomeperperson <= 5000 then Income_Level = "US$ 1,850-5,000"; else if incomeperperson > 5000 and incomeperperson <= 14000 then Income_Level = "US$ 5,000-14,000"; else if incomeperperson > 14000 and incomeperperson then Income_Level = ">US$ 14,000"; else Income_Level='NA'; /*Data Management for lifeexpectancy*/ if lifeexpectancy >= 0 and lifeexpectancy <= 68 then lifeexpectancy_G = " 0-68" ; else if lifeexpectancy > 68 and lifeexpectancy <= 74 then lifeexpectancy_G ="68-74" ; else if lifeexpectancy > 74 then lifeexpectancy_G = "> 76" ; else if lifeexpectancy = . then lifeexpectancy_G = "NA" ; /*Data Management for polityscore*/ if polityscore >= -10 and polityscore <= -4 then polityscore_G = "-10 and -4" ; else if polityscore > -4 and polityscore <= 4 then polityscore_G ="-4 and 4" ; else if polityscore > 4 and polityscore <= 7 then polityscore_G ="4 and 7" ; else if polityscore > 7 and polityscore <= 9 then polityscore_G ="7 and 9" ; else if polityscore > 9 then polityscore_G = "> 9" ; else if polityscore = . then polityscore_G = "NA"; /*Data Management for internetuserrate*/ if internetuserate >= 0 and internetuserate <= 7.5 then internetuserate_G = "0 - 7.5" ; else if internetuserate > 7.5 and internetuserate <= 20 then internetuserate_G ="7.5-20"; else if internetuserate > 20 and internetuserate <= 41 then internetuserate_G = "20-41" ; else if internetuserate > 41 and internetuserate <= 66 then internetuserate_G = "41-66" ; else if internetuserate > 66 then internetuserate_G = ">66" ; else if internetuserate = . then internetuserate_G = "NA"; /*Data Management for breastcancerper100th*/ if breastcancerper100th >= 0 and breastcancerper100th <= 19 then breastcancerper100th_G = " 0 and 19" ; else if breastcancerper100th > 19 and breastcancerper100th <= 26 then breastcancerper100th_G = "19 and 26" ; else if breastcancerper100th > 26 and breastcancerper100th <= 34 then breastcancerper100th_G = "26 and 34" ; else if breastcancerper100th > 34 and breastcancerper100th <= 53 then breastcancerper100th_G = "34 and 53" ; else if breastcancerper100th > 53 then breastcancerper100th_G = ">53" ; else if breastcancerper100th = . then breastcancerper100th_G = "NA"; /*Urban Rate Groups*/ If urbanrate >= 50 then urban=1; else if urbanrate <50 then urban=0; else urban=99;

Proc Sort;by country;Run;

Proc ANOVA;Class urban; Model internetuserate=urban; Means urban; /* urban=1 urbanrate More than or equal to 50% urban=0 urbanrate less than to 50% */

Proc ANOVA; Class polityscore_G; Model internetuserate=polityscore_G; Means polityscore_G/Duncan; where polityscore_G ne 'NA'; Run; /* PolityScore_G groups "Between -10 and -4 points" "Between -4 and 4 points" "Between 4 and 7 points" "Between 7 and 9 points" "Between 9 and points" */

#SAS #ANOVA #statistics

Week 4 - Creating graphs for your data

I will analyse the categorical variables that I have previously created (lifeexpectancy_G, internetuserate_G) as well as the quantitative variable “incomeperperson” using univariate graphs.

As it can be seen from the first chart, life expectancy is slightly left skewed, with around 40% of the observations falling into the category with more than 76 years. Internet Users (per 100 people) we see that almost 20% of observations fall into every category, and this is basically because we use the cumulative frequency distribution to create the groups. Finally, as expected, income distribution is clearly right-skewed, with almost 80% of the observations being on average US$ 6,000.

On average (mean) 36 out of 100 people have access to internet in the data set, with a standard deviation of 28, meaning that this variable has a very high variability. On the other hand, the average life expectancy is 68±10 years.

The hypothesis presented at the beginning was that access to information (expressed as internet penetration) and freedom (polity score) positively contribute to better life conditions (higher life expectancy and less breast cancer). The first analysis is to see if there is are some kind of relationship between each pair of these variables. To do so, I have plotted a scatter graph between internet users (explanatory variable) and life expectancy (response variable). It is clear from the chart below that there might be a positive relationship between these two variables. However, this cannot be interpreted as a causal relationship, since there might be other variables (such us income, freedom, government expenditure) that may be affecting the response variable.

The other pair of variables are breast cancer new cases per 100,000 female and Internet Users. In the graph we can see that there apparently is a positive relationship, which is opposite to what I have proposed in my hypothesis. However this seems to be a weak relationship.

Moreover, I plot a bar chart for Life Expectancy (response variable) and the categorical variable for polityscore i have previously created. The relationship is somewhat consistent with my hypothesis: Greater levels of freedom (higher polityscore) lead to higher life expectancy, except for the countries with the lowest freedom index (-10 to -4).

Finally, below it is presented the codes I have ran in order to get the graph presented in this blog’s entry.

Making Data Management Decisions

I have improved the variables categorisation using the frequency distribution. Variables are now more evenly distributed within each group. The new variables grouping can be seen in the tables below. Income per Person was also transformed as a categoric variable.

Income per capita for instance is distributed in five categories, with about 17 to 18% of observations in each category. One group was created to account for the missing variables, accounting for 10.8% of the countries.

Live expectancy was grouped into 3 categories and one for missing, This variable however seems to be skewed toward ages higher than 70. Number of missing observations are also similar to the previous variable, accounting for 10.33% of the total.

Democracy rating represented by “polityscore” was grouped into five categories as well, and it shows a quite symmetrical distribution; the category that accounts for missing data represents 24.4%.

I only kept the variables (columns) that I shall use in my research.

SAS Code

LIBNAME mydata "/courses/d1406ae5ba27fe300 " access=readonly;

Data Analysis2; Set mydata.gapminder; Label incomeperperson='Income per Capita' HIVrate='Estimated HIV Prevalence %' Internetuserate='Internet users (per 100 people)' breastcancerper100TH='Breast cancer new cases per 100,000 female' lifeexpectancy='Life expectancy at birth (years)' polityscore='Democracy score (Polity)' Income_Level='Income per Capita by Category' breastcancerper100th_G= 'Breast cancer new cases per 100,000 female by Group' lifeexpectancy_G= 'Life expectancy interval' polityscore_G='Polity Score by Group' internetuserate_G='Internet users (per 100 people)by Group'; if incomeperperson >= 0 and incomeperperson <= 560 then Income_Level = "Between 0 and 560" ; else if incomeperperson > 560 and incomeperperson <= 1850 then Income_Level = "Between 560 and 1850" ; else if incomeperperson > 1850 and incomeperperson <= 5000 then Income_Level = "Between 1850 and 5000" ; else if incomeperperson > 5000 and incomeperperson <= 14000 then Income_Level = "Between 5000 and 14000" ; else if incomeperperson > 14000 and incomeperperson then Income_Level = "More than 14,000" ; else Income_Level='NA';

if lifeexpectancy >= 0 and lifeexpectancy <= 68 then lifeexpectancy_G = "Between 0 and 68 years" ; else if lifeexpectancy > 68 and lifeexpectancy <= 74 then lifeexpectancy_G = "Between 68 and 74 years" ; else if lifeexpectancy > 74 then lifeexpectancy_G = " More than 76 years" ; else if lifeexpectancy = . then lifeexpectancy_G = "NA" ;

if polityscore >= -10 and polityscore <= -4 then polityscore_G = "Between -10 and -4 points" ; else if polityscore > -4 and polityscore <= 4 then polityscore_G ="Between -4 and 4 points" ; else if polityscore > 4 and polityscore <= 7 then polityscore_G ="Between 4 and 7 points" ; else if polityscore > 7 and polityscore <= 9 then polityscore_G ="Between 7 and 9 points" ; else if polityscore > 9 then polityscore_G = "Between 9 and points" ; else if polityscore = . then polityscore_G = "NA";

if internetuserate >= 0 and internetuserate <= 7.5 then internetuserate_G = "Between 0 and 7.5 percent" ; else if internetuserate > 7.5 and internetuserate <= 20 then internetuserate_G = "Between 7.5 and 20 percent"; else if internetuserate > 20 and internetuserate <= 41 then internetuserate_G = "Between 20 and 41 percent" ; else if internetuserate > 41 and internetuserate <= 66 then internetuserate_G = "Between 41 and 66 percent" ; else if internetuserate > 66 then internetuserate_G = "Between 66 and percent" ; else if internetuserate = . then internetuserate_G = "NA";

if breastcancerper100th >= 0 and breastcancerper100th <= 19 then breastcancerper100th_G = "Between 0 and 19 new cases per 100,000 female" ; else if breastcancerper100th > 19 and breastcancerper100th <= 26 then breastcancerper100th_G = "Between 19 and 26 new cases per 100,000 female" ; else if breastcancerper100th > 26 and breastcancerper100th <= 34 then breastcancerper100th_G = "Between 26 and 34 new cases per 100,000 female" ; else if breastcancerper100th > 34 and breastcancerper100th <= 53 then breastcancerper100th_G = "Between 34 and 53 new cases per 100,000 female" ; else if breastcancerper100th > 53 then breastcancerper100th_G = "Between 53 and new cases per 100,000 female" ; else if breastcancerper100th = . then breastcancerper100th_G = "NA";

keep incomeperperson breastcancerper100TH HIVrate Internetuserate lifeexpectancy polityscore urbanrate Income_Level lifeexpectancy_G polityscore_G internetuserate_G breastcancerper100th_G;

Proc Freq ; Tables Income_Level lifeexpectancy_G polityscore_G internetuserate_G breastcancerper100th_G; Run;

giuliojcg

Freedom, the Internet and Health

Using the Gapminder I will analyse how access to information through internet can affect peoples’ health. It is important to take into account other variables that may affect access to internet as well, such us income differences. The variables I have selected after a first review are: lifeexpectancy, HIVrate, incomeperperson, Internetuserate, polityscore. I use polityscore as a measure of freedom, HIV rate and life expectancy are the general health measure for a population.

Literature asking similar question have been already published. Internet.Org has found in “Value of Connectivity, Economic and Social Benefits of Internet Access” (2014) that increasing internet penetration in developing countries could boost economic growth, productivity and improvement of health conditions and education opportunities. Stowasser, Heiss, McFadden, and Winter (2011) have conducted an extensive literature review about how socio-economic status (SES) of individuals are positively correlated correlated with their health status. In this study internet access shall be considered one of SES along with other control variables mentioned earlier.

The hypothesis is that, once corrected by income, internet access and freedom are positively correlated to health status measures.

Bibliography:

Deloitte for Facebook Inc. (2014): “Value of Connectivity, Economic and Social Benefits of Internet Access”.

Stowasser, Till, Heiss, Florian, McFadden, Daniel, Winter, Joachim (2011): “Healthy, Wealthy and Wise? Revisited: An Analysis of the Causal Pathways from Socio-economic Status to Health” NBER Working Paper No. 17273.

Exploratory Data Analysis in SAS

LIBNAME mydata "/courses/d1406ae5ba27fe300 " access=readonly;

Data Analysis1; Set mydata.gapminder; Label incomeperperson='Income per Capita' HIVrate='Estimated HIV Prevalence %' Internetuserate='Internet users (per 100 people)' breastcancerper100TH='Breast cancer new cases per 100,000 female' lifeexpectancy='Life expectancy at birth (years)' polityscore='Democracy score (Polity)' Country_G='Polity Score Group' Life_Expectancy_G= 'Life expectancy interval' HIV_Rate_G='HIV rate Range';

/*Create a grouping variable for polityscore*/ if polityscore >=-10 and polityscore <=-8 then Country_G="Between -10 and -8"; else if polityscore > -8 and polityscore <=-5 then Country_G="Between -8 and -5"; else if polityscore > -5 and polityscore <=-2 then Country_G="Between -5 and -2"; else if polityscore > -2 and polityscore <=1 then Country_G="Between -2 and 1"; else if polityscore > 1 and polityscore <=4 then Country_G="Between 1 and 4"; else if polityscore > 4 and polityscore <=7 then Country_G="Between 4 and 7"; else if polityscore > 7 and polityscore <=10 then Country_G="Between 7 and 10"; else if polityscore=. then Country_G="NA";

/*Groups of ages*/ if lifeexpectancy >=0 and lifeexpectancy <=50 then Life_Expectancy_G="Between 0 and 50 years"; else if lifeexpectancy > 50 and lifeexpectancy <=60 then Life_Expectancy_G="Between 50 and 60 years"; else if lifeexpectancy > 60 and lifeexpectancy <=70 then Life_Expectancy_G="Between 60 and 70 years"; else if lifeexpectancy > 70 then Life_Expectancy_G=">70 years"; else if lifeexpectancy=. then Life_Expectancy_G="NA";

/*Groups by GIV rates*/ if HIVrate >=0 and HIVrate <=10 then HIV_Rate_G="Between 0 and 10 percent"; else if HIVrate > 10 and HIVrate <=20 then HIV_Rate_G="Between 10 and 20 percent"; else if HIVrate > 20 then HIV_Rate_G="More than 20percent"; else if HIVrate = . then HIV_Rate_G="NA"; Proc Freq ; Tables Country_G Life_Expectancy_G HIV_Rate_G; Run;

/*Exploratory Analysis of how Income, Internet, and Polityscore vary between each group of Life Expectancy and HIV Rate */ Proc Means Data=Analysis1 Mean Median Max Min STD Missing; /*Label incomeperperson='Income per Capita' HIVrate='Estimated HIV Prevalence %' Internetuserate='Internet users (per 100 people)' lifeexpectancy='Life expectancy at birth (years)' Country_G='Polity Score Group' Life_Expectancy_G= 'Life expectancy interval' HIV_Rate_G='HIV rate Range';*/ Class Country_G; Var incomeperperson Internetuserate lifeexpectancy; Run;

Proc Means Data=Analysis1 Mean Median Max Min STD Missing; Class HIV_Rate_G; Var incomeperperson Internetuserate lifeexpectancy; Run;

Proc Means Data=Analysis1 Mean Median Max Min STD Missing; Class Life_Expectancy_G ; Var incomeperperson Internetuserate lifeexpectancy; Run;

Since most variables in the GAPMINDER data set are continuos and not categorical variables such those in the example video, I had to categorised some of the data of my interest in order to make preliminary analysis of the data.

For the variable “polityscore” I divided the variable into 7 groups, and one that accounts fro missing data. Most countries fall in groups with score going from 4 to 10 (45.54%). The number of countries with no information (missing) represents 24.41% of the total, which is quite significant.

Life Expectancy has been divided into 4 groups and again one more for missing values (10.33%). Almost 54% of the countries in the dataset have a life expectancy of more than 70 years.

The HIV rate variable is the one with more missing values among the selected variables (66 obs. - 30.99% of the total). The number of countries with high HIV rate (more than 20%) are very low, only 3. The presence of a considerable quantity of missing values may present some problems for the data analysis.

Additionally, using the categorical variables I’ve created I’ve used the “Proc Means” statement to explore the data and see if I can find initial evidence of the relationships I am interested in.

Freedom, the Internet and Health

The hypothesis is that, once corrected by income, internet access and freedom are positively correlated to health status measures.

Bibliography:

Deloitte for Facebook Inc. (2014): “Value of Connectivity, Economic and Social Benefits of Internet Access”.