Survey Reveals Diversity in the “Learn to Code” Movement
Do you use any free "learn to code" website to teach yourself programming? You may already know how to program in MATLAB, but you may very well be learning other skills on MOOCs.
Today's guest blogger, Toshi, analyzed a publicly available survey data to understand the demographic of self-taught coders.
Contents
- Load Data
- Higher Female Representation Than Expected
- Mostly Studying in Countries of Citizenship
- Ethnically Diverse English Speakers in US
- Many Are Highly Educated and Already Employed in the US
- Many Already Work In Software Development and IT in US
- Academic Background in Software Development and IT
- Wide Income Gap in Software Development and IT
- What Affects Income in Software Development and IT?
- Age Factor
- Big Companies Not Preferred
- Dream Jobs
- Student Loan Debt
- Women Prefer More Welcoming Venues
- Summary
Load Data
I came across a Techcrunch article, Free Code Camp survey reveals demographics of self-taught coders, and I got curious because a lot of people seem to interested in learning how to code, and industry and government are also encouraging this trend. But programming is hard. Who exactly are the kind of people who have taken the plunge? Our own free interactive online programming classes on MATLAB Academy or gamified MATLAB Cody are also gaining popularity and I would like to understand what motivates this interest.
The survey was conducted anonymously and published on the web and promoted via social media from March 28 through May 2, 2016, targeting people who are relatively new to programming.
The following analysis shows significant diversity in gender and ethnic mix among self-taught coders and the possible impact of MOOCs in opening up access for under-served populations by traditional STEM education paths.
I first downloaded the 2016 New Coder Survey result from Github. I then unzipped the CSV files into my current folder. There are two files - part 1 and part 2 - and we will read them into separate tables. We could perhaps merge them using innerjoin, but in this case I am primarily interested in part 2 only and we will be discarding at least 1000 responses from part 1, given the differences in number of responses.
warning('off','MATLAB:table:ModifiedVarnames') % suppress warning csv = '2016 New Coders Survey Part 1.csv'; % filename part1 = readtable(csv); % read into table part1.Properties.VariableNames = ... % format variable names regexprep(part1.Properties.VariableNames,'_+$',''); % by removing extra "_" csv = '2016 New Coders Part 2.csv'; % filename part2 = readtable(csv); % read into table part2.Properties.VariableNames = ... % format variable names regexprep(part2.Properties.VariableNames,'_+$',''); % by removing extra "_" warning('on','MATLAB:table:ModifiedVarnames') % enable warning part1.SubmitDate_UTC = datetime(part1.SubmitDate_UTC); % convert date strings to datetime part2.SubmitDate_UTC = datetime(part2.SubmitDate_UTC); % convert date strings to datetime s = sprintf('part1 %d responses from %s thru %s\n', ... % summary of part1 height(part1),datestr(min(part1.SubmitDate_UTC)), ... % count of responses, start date datestr(max(part1.SubmitDate_UTC))); % and end date fprintf('%spart2 %d responses from %s thru %s', ... % summary of part2 s,height(part2),datestr(min(part2.SubmitDate_UTC)), ... % count of responses, start date datestr(max(part2.SubmitDate_UTC))); % and end date
part1 15653 responses from 29-Mar-2016 21:23:43 thru 02-May-2016 19:12:44 part2 14625 responses from 29-Mar-2016 21:25:36 thru 02-May-2016 18:35:59
Higher Female Representation Than Expected
Let's start by plotting a histogram of age distrubution. Loren pointed out we can use the omitnan flag in median to deal with missing values instead of nanmedian.
The histogram shows that a lot of people who responded to this survey fall into the so-called "millenials" category. It is interesting to see the number of women who responded to this survey, considering the often cited gender gap in STEM fields. It is not clear if this reflects the true population or are women are over-represented via self-selection? Or somehow self-teaching programming more appealing to women than traditional instruction?
age = part2.HowOldAreYou; % get age from part2 gender = categorical(part2.What_sYourGender); % get gender from part2 as categorical part2.What_sYourGender = gender; % update table figure % new figure x = age(age ~= 0 & gender == 'male'); % subset age by gender histogram(x) % plot histogram text(50,550, sprintf('Median Age (Male) : %d', ... % annotate median(x,'omitnan'))) text(50,470, sprintf('Mode Age (Male) : %d',mode(x))) % annotate hold on % don't overwrite x = age(age ~= 0 & gender == 'female'); % subset age by gender histogram(x) % plot histogram text(50,520, sprintf('Median Age (Female): %d', ... % annotate median(x,'omitnan'))) text(50,440, sprintf('Mode Age (Female) : %d',mode(x))) % annotate hold off % restore default title('Age Distribution by Gender') % add title xlabel('Age') % add x axis label ylabel('Count') % add y axis label legend('Male','Female') % add legend
Mostly Studying in Countries of Citizenship
Since the survey was done online, anyone could participate. Let's check the geographic breakdown. As you would expect, the largest portion of the responses came from the US. You can also see that female responses were 40.59% of male responses in the US, confirming high female representation in the responses.
China is notably missing from the top 10 countries. Perhaps the "learn to code" buzz has not caught on there?
part2.WhichCountryDoYouCurrentlyLiveIn = ... % convert to categorical categorical(part2.WhichCountryDoYouCurrentlyLiveIn); country = part2.WhichCountryDoYouCurrentlyLiveIn; % get country of residence catcount = countcats(country); % get category count cats = categories(country); % get categories [~, rank] = sort(catcount,'descend'); % rank category by count below_top10 = setdiff(cats,cats(rank(1:10))); % categories below top 10 country = mergecats(country, below_top10, 'Other'); % merge them into other country = reordercats(country,[cats(rank(1:10));{'Other'}]);% reorder cats by ranking ratio = sum(country == 'United States of America' & ... % ratio of female/male in us gender == 'female')/sum(country == 'United States of America' & gender == 'male'); figure % new figure histogram(country(gender == 'male')) % plot histogram hold on % don't overwrite histogram(country(gender == 'female')) % plot histogram hold off % restore default ax = gca; % get current axes handle ax.XTickLabelRotation = 90; % rotate x tick label title('Country of Residence by Gender') % add title ylabel('Count') % add y axis label legend('Male','Female') % add legend text(1.5, 1900, sprintf('US Female/Male %.2f%%',ratio*100)) % annotate
You can also visualize migration patterns by mapping countries of citizenship to countries of residence. The number of edges are just 467 - meaning only 467 out of all 14,625 responses in part2 are from migrants, and most people live and study in their countries of citizenship. If you take the ratio of immigration over emigration, US, United Kingdom, Canada, Australia Germany and Russia enjoy net gains from any brain drain.
part2.WhichCountryAreYouACitizenOf = ... % convert to categorical categorical(part2.WhichCountryAreYouACitizenOf); citizenship = part2.WhichCountryAreYouACitizenOf; % get country of citizenship tbl = table(cellstr(citizenship),cellstr(country)); % create table of residence and citizenship tbl(isundefined(citizenship) & isundefined(country),:) = [];% drop empty rows tbl.(1)(strcmp(tbl.(1),'<undefined>')) = ... % use residence if citizenship is emtpy tbl.(2)(strcmp(tbl.(1),'<undefined>')); tbl.(2)(strcmp(tbl.(2),'<undefined>')) = ... % use citizenship if residence is emtpy tbl.(1)(strcmp(tbl.(2),'<undefined>')); [tbl, ~, idx] = unique(tbl,'rows'); % eliminate duplicate rows w = accumarray(idx, 1); % use count of duplicates as weight G = digraph(tbl.(1), tbl.(2), w); % create a directed graph indeg = indegree(G); % get in-degrees ratio = indeg./outdegree(G); % get ratio of in-degrees over out-degrees figure % new figure colormap cool % set colormap w = G.Edges.Weight; % get weights h = plot(G,'MarkerSize',log(indeg+2),'NodeCData',ratio, ... % plot directional graph 'EdgeColor',[.7 .7 .7],'EdgeAlpha',0.3,'LineWidth',10*w/max(w)); caxis([0 3]) % set color axis scaling axis([-2.8 3.3 -4.5 3.7]) % set axis limits title({'Migration Pattern'; ... % add title '467 cases out of 14,625 responses (3.2%)'}) labelnode(h,cats(rank(1:10)),cats(rank(1:10))) % label top 10 nodes nlabels = {'Argentina','Azerbaijan','Chile','Congo', ... % additional nodes to label 'Cote D''Ivoire','Croatia','Greece','Guyana','Latvia','Lesotho', ... 'Malta','Other','Paraguay','Philippines','Republic of Serbia','Romania'}; labelnode(h,nlabels,nlabels); % label additional nodes h = colorbar; % add colorbar ylabel(h, 'in-degrees over out-degree ratio') % add metric
Ethnically Diverse English Speakers in US
Let's focus on the US. As noted earlier, most new self-taught coders who responded to this survey were US citizens, but its ethnic makeup is very diverse. More than half of the women are ethnic minorities, and so are 1/3 of men. They are also predominantly English speakers, given the low ratio of immigrants. However, we should note that the survey itself was in English and promoted via social media in English.
isminority = part2.AreYouAnEthnicMinorityInYourCountry; % get monority status figure % new figure subplot(1,2,1) % create a subplot x = gender(country == 'United States of America' ... % subset & isminority == 0); % us non-minority histogram(x) % plot histogram hold on % don't overwrite x = gender(country == 'United States of America' ... % subset gender & isminority == 1); % by us minority histogram(x) % plot histogram hold off % restore default title('US Gender by Ethnic Category') % add title ylabel('Count') % add y axis label legend('Majority','Minority', 'Location','northwest') % add legend part2.WhichLanguageDoYouYouSpeakAtHomeWithYourFamily = ... % convert to categorical categorical(part2.WhichLanguageDoYouYouSpeakAtHomeWithYourFamily); lanuage = part2.WhichLanguageDoYouYouSpeakAtHomeWithYourFamily;% get language usa = lanuage(country == 'United States of America'); % extract us data catcount = countcats(usa); % get category count cats = categories(usa); % get categories [~, rank] = sort(catcount,'descend'); % rank category by count below_top10 = setdiff(cats,cats(rank(1:10))); % categories below top 10 usa = mergecats(usa, below_top10, 'Other'); % merge them into other usa = reordercats(usa,[cats(rank(1:10)); {'Other'}]); % reorder cats by ranking ax = gca; % get current axes handle ax.XTickLabelRotation = 90; % rotate x tick label subplot(1,2,2) % create a subplot histogram(usa(gender(country == ... % plot histogram 'United States of America') == 'male')) % subset us language by gender hold on % don't overwrite histogram(usa(gender(country == ... % plot histogram 'United States of America') == 'female')) % subset us language by gender hold off % restore default title('US Languages by Gender') % add title ylabel('Count') % add y axis label legend('Male','Female') % add legend
Many Are Highly Educated and Already Employed in the US
We already know that a lot of people who take MOOCs had already earned college degrees and have jobs. This survey also shows the same result.
part2.What_sTheHighestDegreeOrLevelOfSchoolYouHaveCompleted = ... % convert to categorical categorical(part2.What_sTheHighestDegreeOrLevelOfSchoolYouHaveCompleted); degree = part2.What_sTheHighestDegreeOrLevelOfSchoolYouHaveCompleted;% get degree usa = degree(country == 'United States of America'); % extract us data catcount = countcats(usa); % get category count cats = categories(usa); % get categories [~, rank] = sort(catcount,'descend'); % rank category by count usa = reordercats(usa,cats(rank)); % reorder cats by ranking figure % new figure subplot(1,2,1) % create a subplot histogram(usa(gender(country == ... % plot histogram 'United States of America') == 'male')) % subset us degree by gender hold on % don't overwrite histogram(usa(gender(country == ... % plot histogram 'United States of America') == 'female')) % subset us degree by gender hold off % restore default title('US Degrees by Gender') % add title ylabel('Count') % add y axis label legend('Male','Female') % add legend part2.RegardingEmploymentStatus_AreYouCurrently = ... % convert to categorical categorical(part2.RegardingEmploymentStatus_AreYouCurrently); employment = part2.RegardingEmploymentStatus_AreYouCurrently;% get employment other = part2.Other; % get other in employment isstudent = zeros(size(other)); % set up an accumulator fun = @(x,y) ~cellfun(@isempty,strfind(lower(x),y)); % anonymous function handle isstudent(fun(other,'student')) = 1; % flag if 'studnet' is found isstudent(fun(other,'studying')) = 1; % flag if 'studying' is found isstudent(fun(other,'school')) = 1; % flag if 'school' is found isstudent(fun(other,'university')) = 1; % flag if 'university' is found isstudent(fun(other,'degree')) = 1; % flag if 'degree' is found isstudent(fun(other,'phd')) = 1; % flag if 'phd' is found employment(logical(isstudent)) = 'Student'; % update employment usa = employment(country == 'United States of America'); % extract us data catcount = countcats(usa); % get category count cats = categories(usa); % get categories [~, rank] = sort(catcount,'descend'); % rank category by count usa = reordercats(usa,cats(rank)); % reorder cats by ranking subplot(1,2,2) % create a subplot histogram(usa(gender(country == ... % plot histogram 'United States of America') == 'male')) % subset us employment by gender hold on % don't overwrite histogram(usa(gender(country == ... % plot histogram 'United States of America') == 'female')) % subset us employment by gender hold off % restore default title('US Employment by Gender') % add title ylabel('Count') % add y axis label legend('Male','Female') % add legend
Many Already Work In Software Development and IT in US
It turns out that many respondents already work in software development and IT fields and come from very diverse acadamic backgrounds, including both STEM as well as non-STEM subjects. Since the proportion of women tends to be higher in non-STEM majors, this may explain why we see higher than expected female representation in this survey. It appears that female respondents who studied non-STEM majors in undergraduate are now pursuing a career in software development.
Curiously, we also see many computer science majors and they tend to be men. Why are people who already have a computer science background pursuing self-teaching programming? Shouldn't they have learned it in school?
part2.WhichFieldDoYouWorkIn = ... % convert to categorical categorical(part2.WhichFieldDoYouWorkIn); job = part2.WhichFieldDoYouWorkIn; % get job job = mergecats(job, {'software development and IT', ... % merge similar categories 'software development'}); us_job = job(country == 'United States of America'); % extract us data catcount = countcats(us_job ); % get category count cats = categories(us_job); % get categories [~, rank] = sort(catcount,'descend'); % rank category by count below_top10 = setdiff(cats,cats(rank(1:10))); % categories below top 10 us_job = mergecats(us_job, below_top10, 'Other'); % merge them into other us_job = reordercats(us_job,[cats(rank(1:10)); {'Other'}]); % reorder cats by ranking figure % new figure subplot(1,2,1) % create a subplot histogram(us_job(gender(country == ... % plot histogram 'United States of America') == 'male')) % subset us subject by gender hold on % don't overwrite histogram(us_job(gender(country == ... % plot histogram 'United States of America') == 'female')) % subset us subject by gender hold off % restore default title('US Job Field by Gender') % add title ylabel('Count') % add y axis label legend('Male','Female') % add legend part2.WhatWasTheMainSubjectYouStudiedInUniversity = ... % convert to categorical categorical(part2.WhatWasTheMainSubjectYouStudiedInUniversity); major = part2.WhatWasTheMainSubjectYouStudiedInUniversity; % get academic major us_maj = major(country == 'United States of America'); % extract us data catcount = countcats(us_maj); % get category count cats = categories(us_maj); % get categories [~, rank] = sort(catcount,'descend'); % rank category by count below_top10 = setdiff(cats,cats(rank(1:10))); % categories below top 10 us_maj = mergecats(us_maj, below_top10, 'Other'); % merge them into other us_maj = reordercats(us_maj,[cats(rank(1:10)); {'Other'}]); % reorder cats by ranking subplot(1,2,2) % create a subplot histogram(us_maj(gender(country == ... % plot histogram 'United States of America') == 'male')) % subset us subject by gender hold on % don't overwrite histogram(us_maj(gender(country == ... % plot histogram 'United States of America') == 'female')) % subset us subject by gender hold off % restore default title('US Academic Major by Gender') % add title ylabel('Count') % add y axis label legend('Male','Female', 'Location', 'northwest') % add legend
Here is a quick sanity check. Yes, those computer science majors do seem to work in software development and IT, as indicated by the line width of the edge between them.
tbl = table(cellstr(us_maj),cellstr(us_job)); % create table of us major and us job tbl(isundefined(us_maj) | isundefined(us_job) | ... % remove undefined us_maj == 'Other',:)= []; % remove 'Other' from us major [tbl, ~, idx] = unique(tbl,'rows'); % eliminate duplicate rows w = accumarray(idx, 1); % use count of duplicates as weight G = digraph(tbl.(1), tbl.(2), w); % create a directed graph figure % new figure w = G.Edges.Weight; % get weights h = plot(G,'Layout','layered','LineWidth',5*w/max(w)); % plot the directed graph xlim([-2.5 16]) % x-axis limits highlight(h, unique(tbl.(2)),'NodeColor',[.85 .33 .1]) % highlight job nodes title({'US Majors vs. Job Fields'; ... % add title 'Line Width Varies by Frequency'}) text(-2, 2, 'Majors', 'FontWeight','Bold') % annotate text(-2, 1, 'Job Fields', 'FontWeight','Bold') % annotate annotation('arrow',[.2 .2],[.75 .25], 'Color',[0 .45 .75]) % annotate
Academic Background in Software Development and IT
Taking a deep dive in Software Development and IT, we see that not everyone that is a computer science major and has a diverse academic background is represented in the industry, and they are probably in different career tracks based on their background. Women represent a higher proportion in English, Psychology and Other.
us_maj_it = major(country == 'United States of America' ... % subset major by country & job == 'software development and IT'); % and job catcount = countcats(us_maj_it); % get category count cats = categories(us_maj_it); % get categories [~, rank] = sort(catcount,'descend'); % rank category by count below_top10 = setdiff(cats,cats(rank(1:10))); % categories below top 10 us_maj_it = mergecats(us_maj_it, below_top10, 'Other'); % merge them into other us_maj_it = reordercats(us_maj_it,[cats(rank(1:10)); {'Other'}]);% reorder cats by ranking figure % new figure histogram(us_maj_it(gender(country == ... % plot histogram 'United States of America' & ... % subset us subject by job job == 'software development and IT') == 'male')) % and gender hold on % don't overwrite histogram(us_maj_it(gender(country == ... % plot histogram 'United States of America' & ... % subset us subject by job job == 'software development and IT') == 'female')) % and gender hold off % restore default ax = gca; % get current axes handle ax.XTickLabelRotation = 90; % rotate x tick label title('US Software Development and IT - Majors by Gender') % add title ylabel('Count') % add y axis label legend('Male','Female', 'Location', 'northwest') % add legend
Wide Income Gap in Software Development and IT
Let's compare the income range by Job Field using a box plot. The bottom and top of the box represents the first and third quantiles and the middle red line represents the median, and whiskers represent +/- 2.7 standard deviations. Red "+"s show the outliers.
Compared to other job fields, income range in software development and IT has a wide spread (as indicated by the elongnated box shape and longer whisker), meaning there is good upside potential to do better.
income = part2.AboutHowMuchMoneyDidYouMakeLastYear_inUSDollars;% get income income = str2double(income); % convert to numeric income(income == 0) = NaN; % don't count zero us_income = income(country == 'United States of America'); % extract us data figure % new figure boxplot(us_income,us_job) % create a box plot ylim([0 2*10^5]) % set upper limit title('US Income Distribution by Job Field') % add title ax = gca; % get current axes handle ax.XTickLabelRotation = 90; % rotate x tick label ax.YTickLabel = {'$0','$50k','$100k','$150k','$200k'}; % set y tick label ylabel('Annual Income') % add y axis label
What Affects Income in Software Development and IT?
The first factor that may affect the income in software development and IT is academic background. The box plot shows that Computer Science and Electric Engineering give you the most advantage in getting a higher salary. This is probably the motivation behind the self-learning, how-to-code trend - people want to switch to a more lucrativie career path from their current path or advance more quickly within the same industry.
us_income_it = income(country == 'United States of America' ...% subset income by country & job == 'software development and IT'); % and job figure % new figure boxplot(us_income_it,us_maj_it) % create a box plot ylim([0 2*10^5]) % set upper limit title({'US Income Distribution by Major', .... % add title 'in Software Development and IT'}) ax = gca; % get current axes handle ax.XTickLabelRotation = 90; % rotate x tick label ax.YTickLabel = {'$0','$50k','$100k','$150k','$200k'}; % set y tick label ylabel('Annual Income') % add y axis label
Age Factor
Another important factor to consider is age. If you plot the age against income in Software Development and IT, you see a wide income gap among younger Computer Science majors. Some 25 year-olds can be earning $0 vs.$110,000 a year. Income also seems to plateau as you age. You can use fit with the exp2 option to apply a two-term exponential curve to the data so you can see it easily. Perhaps this provides motivation for CS majors to improve their skills and experience as quickly as possible?
us_age_it = age(country == 'United States of America' ... % subset age by country & job == 'software development and IT'); % and job X = us_age_it(us_maj_it == 'Computer Science'); % subset just CS Y = us_income_it(us_maj_it == 'Computer Science'); % subset just CS figure % new figure plot(X, Y,'o') % plot data hold on % don't overwrite missingrows = isnan(X) | isnan(Y); % find NaNs X(missingrows) = []; % remove NaNs Y(missingrows) = []; % remove NaNs fitresult = fit(X,Y,'exp2'); % fit to exp2 plot(fitresult) % plot curve hold off % restore default title({'Income by Age Among CS Majors'; ... % add title 'in Software Development and IT'}) xlim([10 60]) % set x axis limits ylim([0 2*10^5]) % set y axis limits ax = gca; % get current axes handle ax.YTick = 0:50000:200000; % set y tick ax.YTickLabel = {'$0','$50k','$100k','$150k','$200k'}; % set y tick label xlabel('Age') % add x axis label ylabel('Annual Income') % add y axis label
Big Companies Not Preferred
If you look at the future employment those people are looking for, they prefer big companies the least. Since people are not earning a formal degree, they are probably expecting more flexibility in other employment options.
The most popular option is mid-sized companies, but many people are interested in working for a startup, or starting their own businesses, or freelancing. People in software development and IT tend to prefer working for startup or mid-sized companies more. Men tend to prefer doing their own businesses or work for a startup, while women tend to prefer the freelance path.
part2.want_employment_type = ... % convert to categorical categorical(part2.want_employment_type); interested_emp = part2.want_employment_type; % get interested employment type figure % new figure subplot(1,2,1) % create a subplot us_int_emp_it = interested_emp(country == ... % subset it by country and job 'United States of America' & job == 'software development and IT'); us_int_emp_it(isundefined(us_int_emp_it)) = []; % remove undefined us_int_emp_it = removecats(us_int_emp_it); % remove unused categories histogram(us_int_emp_it,'Normalization','probability') % plot histogram hold on % don't overwrite us_int_emp_non_it = interested_emp(country == ... % subset it by country and job 'United States of America' & job ~= 'software development and IT'); us_int_emp_non_it(isundefined(us_int_emp_non_it)) = []; % remove undefined us_int_emp_non_it = removecats(us_int_emp_non_it); % remove unused categories histogram(us_int_emp_non_it,'Normalization','probability') % plot histogram hold off % restore default title({'US Desired Employment Type';'by Job Field'}) % add title ax = gca; % get current axes handle ax.XTickLabelRotation = 90; % rotate x tick label ax.YTick = 0:0.1:0.6; % set y tick ax.YTickLabel = {'0%','10%','20%','30%','40%','50%','60%'}; % set y tick label legend('Software Dev and IT', 'Others') % add legend ylim([0 0.6]) % set y axis limits subplot(1,2,2) % create a subplot us_int_emp_m = interested_emp(country == ... % subset it by country 'United States of America' & gender == 'male'); % gender us_int_emp_m(isundefined(us_int_emp_m)) = []; % remove undefined us_int_emp_m = removecats(us_int_emp_m); % remove unused categories histogram(us_int_emp_m,'Normalization','probability') % plot histogram hold on % don't overwrite us_int_emp_f = interested_emp(country == ... % subset it by country 'United States of America' & gender == 'female'); % gender us_int_emp_f(isundefined(us_int_emp_f)) = []; % remove undefined us_int_emp_f = removecats(us_int_emp_f); % remove unused categories histogram(us_int_emp_f,'Normalization','probability') % plot histogram hold off % restore default title({'US Desired Employment Type';'by Gender'}) % add title ax = gca; % get current axes handle ax.XTickLabelRotation = 90; % rotate x tick label ax.YTick = 0:0.1:0.6; % set y tick ax.YTickLabel = {'0%','10%','20%','30%','40%','50%','60%'}; % set y tick label legend('Male', 'Female') % add legend ylim([0 0.6]) % set y axis limits
Dream Jobs
When it comes to actual jobs people are interested in, they are mostly web development positions. People already in software development and IT tend to prefer roles with higher technical skills - Back-End Web Development, DevOps, or SysAdmin rather than Front-End Web Development, or other non-development roles such as Product Manager or QA Engineer. In terms of gender, women tend to prefer Front-End Web Development and User Experience Design.
int_job = categorical(strtrim(part2.jobs_interested_in)); % get interested job catcount = countcats(int_job); % get category count cats = categories(int_job); % get categories [~, rank] = sort(catcount,'descend'); % rank category by count below_top10 = setdiff(cats,cats(rank(1:10))); % categories below top 10 int_job = mergecats(int_job, below_top10,'Other'); % merge them into other int_job = reordercats(int_job,[cats(rank(1:10));{'Other'}]);% reorder cats by ranking figure % new figure subplot(1,2,1) % create a subplot us_int_job_non_it = int_job(country == ... % subset int job by country and job 'United States of America' & job ~= 'software development and IT'); histogram(us_int_job_non_it, 'Normalization','probability') % plot histogram hold on % don't overwrite us_int_job_it = int_job(country == ... % subset int job by country and job 'United States of America' & job == 'software development and IT'); histogram(us_int_job_it, 'Normalization','probability') % plot histogram hold off % restore default title({'US Jobs Interested In';'By Job Field'}) % add title legend('All Others','Software Dev & IT') % add legend ax = gca; % get current axes handle ax.YTick = 0:0.1:0.5; % set y tick ax.YTickLabel = {'0%','10%','20%','30%','40%','50%'}; % set y tick label ylim([0 0.5]) % set y axis limits subplot(1,2,2) % create a subplot us_int_job_m = int_job(country == ... % subset int job by country 'United States of America' & gender == 'male'); % and gender histogram(us_int_job_m, 'Normalization','probability') % plot histogram hold on % don't overwrite us_int_job_f = int_job(country == ... % subset int job by country 'United States of America' & gender == 'female'); % and gender histogram(us_int_job_f, 'Normalization','probability') % plot histogram hold off % restore default title({'US Jobs Interested In';'By Gender'}) % add title legend('Male','Female') % add legend ax = gca; % get current axes handle ax.YTick = 0:0.1:0.5; % set y tick ax.YTickLabel = {'0%','10%','20%','30%','40%','50%'}; % set y tick label ylim([0 0.5]) % set y axis limits
Student Loan Debt
The survey also answers how much student loan debt respondents carry and how much they spend learning to code. Over 41% of the respondents have student loan debt and the median amount owed is $25,000. In addition, people do spend during the course of learning to code, and the median total spend is $300. Given that a lot of people have debt, they cannot afford to spend more and add to their deficit, and that also reflects on the more conservative choices in future employment.
debt = part2.AboutHowMuchDoYouOweInStudentLoans_inUSDollars;% get student load debt debt = str2double(debt); % convert to numeric debt(debt == 0) = NaN; % don't count zero us_debt = debt(country == 'United States of America'); % extract us data pct_in_debt = sum(~isnan(us_debt))/length(us_debt)*100; % percentage in debt median_debt = nanmedian(us_debt)/1000; % median debt figure % new figure subplot(1,2,1) % create a subplot histogram(us_debt) % plot histogram xlim([0 2*10^5]) % set y axis limits ax = gca; % get current axes handle ax.XTick = 0:50000:200000; % set x tick ax.XTickLabel = {'$0','$50k','$100k','$150k','$200k'}; % set x tick label xlabel('Amount Owed') % add x axis label ylabel('Count') % add y axis label title({'US Student Loan Debt'; ... % add title sprintf('%.2f%% in Debt (Median $%dk)',pct_in_debt,median_debt)}) subplot(1,2,2) % create a subplot spend = part2.total_spent_learning; % get total spend spend = str2double(spend); % convert to numeric spend(spend == 0) = NaN; % don't count zero us_spend = spend(country == 'United States of America'); % extract us data histogram(us_spend) % plot histogram xlim([0 3*10^4]) % set y axis limits ax = gca; % get current axes handle ax.XTick = 0:10000:30000; % set x tick ax.XTickLabel = {'$0','$10k','$20k','$30k'}; % set x tick label xlabel('Total Spend') % add x axis label ylabel('Count') % add y axis label title({'US Spend Learning'; sprintf('Median $%d', ... % add title nanmedian(us_spend))})
Women Prefer More Welcoming Venues
This survey seems to show more female participation in the "learn to code" movement compared to a more tranditional computer science education. When you look at the type of events women prefer, they show strong preference for gender-specific events like "Girl Develop It" and "Women Who Code". When you look at the online resources, you don't see much difference by gender. It appears that physical presence of males makes women feel unwelcome.
events_attended = part2.attended_event_types; % get events attended events_attended = cellfun(@(x) strsplit(x,','), ... % split by comma events_attended,'UniformOutput',false); events_attended_flatten = strtrim([events_attended{:}]); % un-nest and trim [~,ia,ib] = unique(lower(events_attended_flatten)); % get indices of uniques events = events_attended_flatten(ia); % get unique values count = accumarray(ib,1); % count unique values events(count < 100) = []; % drop unpopular events events(strcmpi(events,'none')) = []; % drop 'none' events(cellfun(@isempty,events)) = []; % drop empty cell attended = zeros(size(events_attended,1),length(events)); % set up accumulator for i = 1:size(events_attended,1) % loop over events attended attended(i,:) = ismember(events,strtrim( ... % find intersection between events_attended{i})); % events and attended events end attended_m = sum(attended(country == ... % subset attended by country 'United States of America' & gender == 'male',:)); % and gender attended_f = sum(attended(country == ... % subset attended by country 'United States of America' & gender == 'female',:)); % and gender gender_ratio = attended_m ./ sum(attended_m); % get male ratio by event gender_ratio = [gender_ratio; attended_f./sum(attended_f)]; % add female ratio figure % new figure subplot(1,2,1) % create a subplot b = bar(gender_ratio','FaceColor',[0 .45 .75], ... % create a bar chart 'FaceAlpha',.6); % with histogram colors b(2).FaceColor = [.85 .33 .1]; % with histogram colors ax = gca; % get current axes handle ax.XTickLabel = events; % set x tick label ax.XTickLabelRotation = 90; % rotate x tick label ax.YTick = 0:0.1:0.4; % set y tick ax.YTickLabel = {'0%','10%','20%','30%','40%'}; % set y tick label title('US Popular Events Attended') % add title legend('Male','Female') % add legend subplot(1,2,2) % create a subplot resources_used = part2.learning_resources; % get resources used resources_used = cellfun(@(x) strsplit(x,','), ... % split by comma resources_used,'UniformOutput',false); resources_used_flatten = strtrim([resources_used{:}]); % un-nest and trim [~,ia,ib] = unique(lower(resources_used_flatten)); % get indices of uniques resources = resources_used_flatten(ia); % get unique values count = accumarray(ib,1); % count unique values resources(count < 100) = []; % drop unpopular resources resources(cellfun(@isempty,resources)) = []; % drop empty cell usage = zeros(size(resources_used,1),length(resources)); % set up accumulator for i = 1:size(resources_used,1) % loop over resources used usage(i,:) = ismember(resources,strtrim( ... % find intersection between resources_used{i})); % resources and resource used end usage_m = sum(usage(country == ... % subset usage by country 'United States of America' & gender == 'male',:)); % and gender usage_f = sum(usage(country == ... % subset usage by country 'United States of America' & gender == 'female',:)); % and gender gender_ratio = usage_m ./ sum(usage_m); % get male ratio by resource gender_ratio = [gender_ratio; usage_f ./ sum(usage_f)]; % add female ratio b = bar(gender_ratio','FaceColor',[0 .45 .75], ... % create a bar chart 'FaceAlpha',.6); % with histogram colors b(2).FaceColor = [.85 .33 .1]; % with histogram colors ax = gca; % get current axes handle ax.XTickLabel = resources; % set x tick label ax.XTickLabelRotation = 90; % rotate x tick label ax.YTick = 0:0.1:0.3; % set y tick ax.YTickLabel = {'0%','10%','20%','30%'}; % set y tick label title('US Popular Resources Used') % add title legend('Male','Female') % add legend
To give an example, I encouraged my daughter to join a robotics competition team in her high school. She talked to her friends because she didn't want to be only girl in the team and a bunch of girls joined the team. When she came home from the first team session, I asked her what she worked on. She said "we worked on team web page". It turned out boys worked on the building robots and girls were left out, so they worked on building the team web page. When the kits were delivered to the team, boys just huddled togather among themselves, and didn't bother to include girls. Girls were not consciously excluded, but they felt unwelcome anyway. I suspect similar dynamics may be at play which coding events women go to.
I also wonder the female preference of Front-End Web Development and User Experience Design is also driven by the same issue?
Summary
Perhaps the most intriguing result of this analysis is that the "learn to code" movement is effective in closing the gender gap in software development and IT and embraced by the minority community under-served by the traditional educational paths. It also underscores the precarious positions those learners face due to the high student loan debt they carry. Ultimately we don't know how many of them actually achieve employment in their dream job from this survey, and hopefully there is a follow-up to find out whether the "learn to code" movement really delivers on its promise.
Do you use any of those "learn to code" websites or other MOOCs? What are you learning and what motivates you to take those classes? Please share your experience here!
- Category:
- Social Computing,
- Statistics
