Pay results fee

Sorry, you have not enough credits to pay results fee.

# Improved measurement of behaviour change within the YAL community

Status Closed Closed 2 years ago
Bounty Reward
50

### Labels

Statistics Design of experiments Statistical modelling

### Description

We run an edutainment HIV/AIDS portal called Young Africa Live in South Africa, Kenya and Tanzania. For more information please look at our website http://praekeltfoundation.org/young-africa-live.html.

The primary objective of what we're trying to do is to increase sexually safe and responsible behaviour in youths age 18 to 25. In addition we aim to drive positive change in the preconceptions about gender and sexuality in this audience. For the past two years we've conducted surveys on the platform to determine our users' opinion on relationship and sexual practices. The results can be seen in these two PDFs:

Additionally, we record web analytics of user behaviour on the platform, including page impressions, number of comments posted and top articles that people have viewed. We have dashboards that show an overview of these analytics, an example dashboard for Tanzania can be seen here. Dashboards are also available for South Africa and Kenya. It may be possible to source more in-depth data if required by a data scientist.

We are looking for suggestions about how we can improve our practices to monitor change in user behaviour over time. We're also looking for additional survey questions or data that we should be collecting that would assist us in achieving this objective, as well as any suggestions for equating online measurements with real world behaviours.

In the future we're looking to employ a statistician to help analyse our data and make ongoing suggestions about improvements that we could be making. The best answers from this "bounty" will help us determine who we'd like to employ. If you require further information or need clarification on anything, feel free to post a clarification and we'll do our best to answer it.

• Feb 08, 2013

Sex Survey: Q3 - what about a guy who thinks sex is better with love? What about a girl who just likes sex? Q6 - Categories overlap, you have 16-18 then 18-21 which category should someone answer if they were 18? You will get unreliable results Q12 - again you are asking two questions here. How can someone answer this if they know someone who has? Surely you still want to know if their answer for themself is yes or no? Q13 - answer might be yes and they are not ina relationship now but it didn't end badly. Q15 - this is not an options questions there are so many things someone might say, just allow open ended response Q18 - you have stated 'if nothing physical happens' so then having an option of 'only if you have sex' doesn't make sense Q29 - what about i kept my baby and it was the wrong decision? Might be a sad answer but it should be an option

I have not listed all the possible issues with this questionnaire too as I think it is important to go on to the analysis side of things, however I think both questionnaires need quite a lot of revision in order to make your results reliable.

In terms of analysis of course you can just look at the % change over time, probably split into males and females and age groups, but it would also be interesting to go into more depth on some of the questions. At the moment I think you might be asking a few too many questions and so it would be hard to really get a focus on the issues. Although the blog/news items questions are interesting these will be changing over time and so won't be comparable and so maybe cut down on these.

I think the dashboard could show more useful data, currently it is really just showing you the amount of traffic through the polls which may be useful but doesn't give you any overview of the data.

In terms of analysis I think you should be looking at running association tests, chi squared being the most basic, to see if there are relationships between people views in different areas. What questions you would look at would depend on more detail about what you want to find out. In the future I think it would be good to get some scale questions in there [what is your opinion on a scale of 1-5 etc] largely because this gives much more scope for statistical analysis and you would be able to build models around what is predicting people's views on sex/HIV and so on. In the same vein as this, asking for people's actual age rather than categories would be good as then this could be used as a predictor and we could also run ANOVAs to look at whether views differ significantly depending on age. There are many directions in which this could go, and I hope to be able to work with you on this to help develop such an important project.

• Feb 05, 2013

Hi,

I have looked into your survey. It would be better if the survey is shorter (less questions) and returning webpage visitors (with user account) can be ensured they will get new set of questions, to keep them in active participation. This also allows changes in opinion on sexual practices and AIDS be recorded and correlations over time be captured. Correlations between answers to different questions are important to understand their perceptions on HIV/AIDS, for example, to understand how they would think and react if faced a certain social situation. Normal statistics using averages does not capture them. If this is expensive, one can still use the raw data obtained from each questions and analyze their overall correlations, but it does not have the possibility to re-test/verify the conclusions obtained from the analysis, since it is a one-time only survey. These would be my suggestions.

• Feb 05, 2013

• Feb 07, 2013

Hi Belinda, I read your interested work. I’ have some suggestion on the question present in the questionnaire. The first is general, try to not put comment in the possible answer, because it could to condition the responder, I mean that someone can try to search the “right” answer and not his answer. In question 3 the possible answers are separately between girl and guy but if you have gender in the interview you don’t need to specify and some guy can responder that he like to be involved, otherwise if the first answer is “ I’m girl …” no guy could say the same … Question 5 yes no I don’t’ know are the perfect answer without comment, because the first is “yes, there is no other way …” and I think Yes but there no the only way? Question 6 don’t’ put comment as too young or “!” after I’m virgin it’s easier to have good answer in this way. In some question like Question 23 and 24 I think it could be better to can choice more options, to do this you don’t need to create combination, only you can calculate frequency in another way. Generally I think that could be interested analyzed data by sex and classes of age to understand which is the subgroups to make more interventions. The questionnaire is long I suggest to create a fixed part to present every year and another part that you repeat every 3 years. So the not fixed part could be changed and you can make some question in different period. I don’t understand if there are some demographic question and if yes which are. Could be interested know if this girls and guy work or study or the socioeconomic status, religion. Also you can insert one question like: DO you think that young people need more advice on HIV? Yes no If yes which the best way to receive this advice for you?

Have good luck

• Feb 08, 2013

Based on the program description, I thought about implementing the following: 1. Weight the questions i.e. assign a numerical value to each answer to the questions based on its correctness/near correctness. Create clusters of points that can be earned. Total cumulative points earned by each participant and place the participants in their respective clusters. 2.Correlate the questions. For example, answer A to question 1 should lead to answer B to question 2. Analyze if the participant acted in accordance to this rule. It is then easier to deduce behavior. 3.Let them retake the test after sometime. Has there been a shift in their opinion to the questions? Use innovative technologies such as knowledge management and information retrieval and machine learning to retrieve information about each participant. 4.Use clustering to put students with similar characteristics together. Using these identify any other features they share. 5.Use of cutting edge data mining, machine learning techniques for building advanced supervised learning systems. These are systems predict future behaviors based on some observed information.

• Feb 15, 2013

I have read the report on the SA Youth sex survey 2011 and the YAL 2012 . I can see that they have collected some meaningful data, however, I have a number of fundamental concerns about the results presented.

*  age and gender details by region of respondents were not supplied,
*  proportions of questions answered  by age and gender and region,

* where are the margin of error estimates  - precision estimates are required at a minimum.


This type of survey should be used to monitor behavioural change as a result of national/local policy intervention programmes on areas such as HIV/AIDS, safe sex and behaviours to women and other minority groups.

However, they do need to urgently address the issue I have raised in relation to representativeness and precision.

• AlixGodfrey said:

Hi,

I just wanted to say thanks for choosing my answer and hope to work with you in the future.

If you do want to contact me about future projects and analysis feel free to email me at alix.godfrey@sigmaphi.co.uk, although happy to work through Zombal.