Difference between revisions of "SurveyCTO Programming Work Flow"
Revision as of 23:14, 13 February 2017
While it is possible to code a SurveyCTO questionnaire in many different ways, the time it will take and the quality of final product depends a lot on how the work is planned. This article outlines an approach common among experienced SurveyCTO programmers at DIME.
- Do not start at the top of the survey and work your way through one question at the time. That takes longer and tends to end up in less well structured questionnaires that are much more error prone.
All surveys should always start with creating a paper questionnaire. See Questionnaire Design and Survey Pilot for instructions on how to develop and test a paper based survey. When there is no time create and pilot the paper based version of the questionnaire, a questionnaire should still be created in Word or Excel before starting to code the survey in SurveyCTO code. Skipping this step almost always leads to issues with data quality.
The important part of doing a paper questionnaire before starting to code is that it allows the project team to focus on what question to ask and how to ask them. The questionnaire coding can be implemented to any format, display and functionality decided on in the paper questionnaire. By focusing on content first and programming implementation only afterwards, the quality of the data we collect gets a higher priority than when the questionnaire is set up in a way which is technically convenient to program. If the SurveyCTO code is implemented before or in parallel to developing the content, then the code always takes focus away from the questionnaire and negatively impacts the way a question is asked and answered.
Writing pseudo code is a method that all computer scientists use, before starting to write code, when they are not exactly sure how all details of that code will be in the end. It is very rare that anyone starting to code a SurveyCTO questionnaire knows how all modules will be coded, so writing pseudo code is the best place to start.
Writing pseudo code is simple to start. Start by describing at a high level what the code will do (often in bullet points) in plain English. After all functions of the code has been described at a high level, more details are added. As details are added, the language also changes from plain English to something that resembles code more and more. Computer scientist very often write pseudo code with pen and paper or on a white board.
The next subsection gives a quick introduction on how to write pseudo code in relation to a SurveyCTO questionnaire.
Listing the modules
The highest level a questionnaire can be described on are the modules. So start by writing a list of the modules in your questionnaire. Write them in the order that you want them to be asked:
|1. Pseudo Code|
Adding groups and repeat groups
The second highest level of detail is adding all the groups and repeat groups that will control the flow of the questionnaire. Some groups are used to change appearances of the questions, and those group does not need to be listed here. The groups important to list here are the groups that are used as skip patterns.
When you add a repeat group or a group used as a skip pattern, also add what will determine the number of repeats or what will cause the group to be skipped. We are still writing pseudo code so it is perfectly fine to write in plain English, although some code is also good.
|2.1 Pseudo Code|
If you look at the pseudo code above, you will see that the repeat in the employment module depends on a variable that is not asked until the household roster. This is the purpose of writing pseudo code. We want to, as early as possible, catch anything that will cause a problem while we are still working on a high level, and before we have spent a lot of time working on details. If we first would have spent a week working out the detail of one module, and later figure out that we need to make a change to the structure of the questionnaire, then there is a risk that we have to re-do that work.
|2.2 Pseudo Code|
Adding more specific references
From here on, it is a iterative process of adding a few details at the time. Focus on adding repeat groups, groups with skip patterns and any variables that they reference. Also add references such as being able to include labels from previous modules. For example, in the pseudo code below we most likely want to be able to reference the names listed in the household roster in the employment section. We might also want to be able to have all women aged 16-45 listed as answer options in any question in the Access to Maternity Health Care section. Repeat this step and add a bit more detail in each repetition, and each time make sure that the structure still suits what you intend to do.
|3. Pseudo Code|
When is Pseudo Code Complete?
There is no clear rule for when a pseudo code is completed, but the more experienced a coder gets the more detail is included in their pseudo code. For a beginner it might be difficult to predict what added detail will require an update to the structure the more detailed it gets. It is possible to take a break in pseudo coding and code up the structure that you have in the pseudo code already. Do some experimentation to test logic that you are unsure of, and then return back to working on your pseudo code.
The point of pseudo code is to get an idea of the details that are needed for the flow of the questionnaire. When you have your structure (sometimes called "questionnaire skeleton") then you can move on to implementing your pseudo code in actual SurveyCTO code.
The more time you spend on pseudo coding the easier the actual coding will be. Spend some time for at least a few days on this. Write up your first pseudo code and then sleep on it. Review it with fresh eyes the next day. Or even better, when you have a final draft, discuss your pseudo code with someone else in their project. Proofreading pseudo code is just as useful as asking someone to proofread your text documents.
You have now reached the point when you can start coding. If you both have first created a paper survey and written your pseudo code, then you to the best degree possible have reduced the risk of running in to time consuming major issues while programming. However, there are often unexpected issues that even the most experienced coders run in to. In those cases, it might be helpful to go back to the paper survey or the pseudo code and see if you can come up with a solution that way.
There are two ways to do programming in SurveyCTO. Most programmers that start by developing pseudo code usually find it easiest to implement pseudo code using the Excel method. Since, we also strongly recommend everyone writing large household surveys to use the Excel method, we will assume that through the rest of this article.
When you start coding, never start with an empty excel sheet. SurveyCTO provides template forms that have the basic settings already in place. There is a template with the just the basic settings. That is what most programmers usually start with. The other forms are also used a lot, but maybe more often to used as inspiration or code examples that are copied and pasted into your questionnaire.
To create a new form using a form template, log in to your server and go to the design tab. Then click
start new right at the top of the section Your Forms. After giving the form a name and an id (remember to be plan ahead when creating the id), you can select which form template that you want to use. If this is a form for which you eventually will collect real data on, remember to encrypt your form.
What to program first?
When you program, do not start with the first question and program your way to the last question. You always want to code from high level to small detail. This is what you have prepared for in your pseudo code, but even if you do not have pseudo code, you should still start with high level and work your self to details.
Start with creating the repeat groups and skip groups that you have listed in your pseudo code. Then create the variables that the repeat counts and skip patterns depend on. Then make sure that there are at least one variable in each repeat group and skip group. After that is completed, upload your survey and test you code thoroughly for the first time. Even if you have spent a lot of time on your pseudo code, you still want to test that all your assumed programming logic works as intended before you spend time working on details.
Filling in the details
When you have tested your structure, then you can start to fill in the details module by module. Start with the module that includes the most complex thing you want to implement, but if that module depends on any other module, then you must do that module first. In each module, keep approaching your code in the same way as before, i.e. from high level to detail.
Back to parent topic
This article is a part of the topic Questionnaire Programming.