Seantis Questionnaire

Introduction

Seantis Questionnaire is a django questionnaire app which is easily customised and includes advanced dependency support using boolean expressions.

It allows an administrator to create and edit questionnaires in the django admin interface, with support for multiple languages.

It was originally created to support an annually recurring questionnnaire for a medical study.

This repository only contains the Questionnaire side of the application. We also developed a management interface and extensions to the models that are specific to the study, and have therefore not been made public. Seantis GmbH could, however, provide a similar end-to-end solution for your organisation.

Alternatives

There are two other questionnaire-type applications that we stumbled upon, but they both didn't quite scratch our itch, but they may scratch yours.

Django Questionnaire - http://djangoquest.aperte-it.com/

Django Survey - http://code.google.com/p/django-survey/

History

The questionnaire app was originally developed by rmt for Seantis. We picked up the project again in 2011 for yet another medical study. At this point we decided to introduce features and changes that break backwards compatibility with existing questionnaires that use this app.

The old versions are tagged as follows:

• tag 1.0 - state of last commit by the original developer (rmt)
• tag 1.1 - contains merged changes by other forks improving the orignal

The new version is the current trunk and is dubbed v2.0.

It is still possible to use a questionnaire defined for 1.0 and use it in 2.0 and we cover that topic later in this manual. However, if you used 1.0 and integrated it into your own page you are on your own. We only cover data-migration from an existing questionnaire 1.0 instance to a new questionnaire 2.0 instance.

A lot of template code has changed from 1.0 to 2.0 because we switched from the blueprint CSS framework to the Twitter bootstrap CSS framework. There is also some cleanup around the urls and static files. That's why moving to 2.0 might be a somewhat hairy task to accomplish.

About this Manual

Seantis Questionnaire is not a very well documented app so far to say the least. This manual should give you a general idea of the layout and concepts of it, but it is not as comprehensive as it should be.

What it does cover is the following:

• Integration talks lays out the steps needed to create a new Django page together with the questionnaire. The same steps can be used to integrate the questionnaire into an existing site (though you would be entering unpaved ways).
• Conecpts talks about the data model and the design of the application.
• Migration explains how a questionnaire defined with 1.0 can be used in 2.0.
• 2.0 Postmortem talks about some experiences made during the development of 2.0.

Integration

This part of the docs will take you through the steps needed to create a questionnaire app from scratch. It should also be quite handy for the task of integrating the questionnaire into an existing site.

First, create a folder for your new site:

mkdir site
cd site

Create a virtual environment so your python packages don't influence your system

virtualenv --no-site-packages -p python2.5 .

Activate your virtual environment

source bin/activate

Install Django

pip install django

Create your Django site

django-admin.py startproject mysite

Create a place for the questionnare

cd mysite
mkdir apps

Clone the questionnaire source

git clone git://github.com/seantis/seantis-questionnaire.git

You should now have a seantis-questionnaire folder in your apps folder

cd seantis-questionnaire

The next step is to install the questionnaire.

python setup.py install

If you are working with seantis-questionnaire from your own fork you may want to use python setup.py develop instead, which will save you from running python setup.py install every time the questionnaire changes

Next up we'll have a look at configuring your basic questionnaire.

First, you want to setup the languages used in your questionnaire, by opening settings.py in your site's folder (the one with the subfoler apps/)

Open settings.py and add following lines, representing your languages of choice:

LANGUAGES = (
    ('en', 'English'),
    ('de', 'Deutsch')
)

At the top of settings.py you should at this point add

import os.path

We will use that below for the setup of the folders

In the same file add the questionnaire static directory to your STATICFILES_DIRS

os.path.abspath('./apps/seantis-questionnaire/questionnaire/static/')

Also add the locale and request cache middleware to MIDDLEWARE_CLASSES

'django.middleware.locale.LocaleMiddleware',
'questionnaire.request_cache.RequestCacheMiddleware',

Add the questionnaire template dir as well as your own to TEMPLATE_DIRS

os.path.abspath('./apps/seantis-questionnaire/questionnaire/templates'),
os.path.abspath('./templates'),

And finally, add transmeta, questionnaire and django.contrib.markup to your INSTALLED_APPS

'transmeta',
'questionnaire',
'questionnaire.page',
'django.contrib.markup',

Next up we want to edit the urls.py of your project to hookup the questionnaire views with your site's url configuration.

For an empty site with enabled admin interface you should end up with something like this:

from django.conf.urls.defaults import patterns, include, url

from django.contrib import admin
admin.autodiscover()

urlpatterns = patterns('',
    url(r'^admin/', include(admin.site.urls)),
    
    # questionnaire urls
    url(r'q/', include('questionnaire.urls')),
    
    url(r'^take/(?P<questionnaire_id>[0-9]+)/$', 'questionnaire.views.generate_run'),
    url(r'^$', 'questionnaire.page.views.page', {'page' : 'index'}),
    url(r'^(?P<page>.*)\.html$', 'questionnaire.page.views.page'),
    url(r'^(?P<lang>..)/(?P<page>.*)\.html$', 'questionnaire.page.views.langpage'),
    url(r'^setlang/$', 'questionnaire.views.set_language'),
)

For the questionnaire itself it is only necessary to have the urls below `# questionnaire urls

Having done that we can initialize our database. For this to work you must have setup your DATABASES in settings.py

python manage.py syncdb

The questionnaire expectes a base.html template to be there, with certain stylesheets and blocks inside. Have a look at ./apps/seantis-questionnaire/example/templates/base.html

For now you might want to just copy the base.html to your own template folder.

mkdir templates
cd templates
cp ../apps/seantis-questionnaire/example/templates/base.html .

Congratulations, you have setup the basics of the questionnaire! At this point this site doesn't really do anything, as there are no questionnaires defined.

To see an example questionnaire you can do the following (unfortunately, this will only work if you have both English and German defined as Languages in settings.py)

python manage.py loaddata ./apps/seantis-questionnaire/example/fixtures/initial_data.yaml

You may then start your development server

python manage.py runserver

And navigate your browser to localhost:8000

Concepts

The Seantis Questionnaire sports the following tables, described in detail below.

• Subject
• RunInfo
• RunInfoHistory
• Question
• Choice
• QuestionSet
• Questionnaire
• Answer

Subject

A subject is someone filling out a questionnaire.

Subjects are primarily useful in a study where the participants answer a questionnaire repeatedly. In this case a subject may be entered. Whoever is conducting the study (i.e. the person running the questionnaire app), may then periodically send emails inviting the subjects to fill out the questionnaire.

Sending Emails is covered in detail later.

Of course, not every questionnaire is part of a study. Sometimes you just want to find out what people regard as more awesome: pirates or ninjas*?

*(it's pirates!)

Though a poll would be a better choice for this example, one can find the answer to that question with Seantis Questionnaire by using an anonymous subject. The next chapter Questionnaire will talk about that in more detail.

RunInfo

A runinfo refers to the currently active run of a subject.

A subject who is presently taking a questionnaire is considered to be on a run. The runinfo refers to that run and carries information about it.

The most important information associated with a runinfo is the subject, a random value that is used to generate the unique url to the questionnaire, the result of already answered questions and the progress.

Once a run is over it is deleted with some information being carried over to the RunInfoHistory.

Runs can be tagged by any number of comma separated tags. If tags are used, questions can be made to only show up if the given tag is part of the RunInfo.

RunInfoHistory

The runinfo history is used to refer to a set of answers.

Question

A question is anything you want to ask a subject. Since this is usually not limited to yes or no type questions there are a number of different types you can use:

• choice-yesno - Yes or No
• choice-yesnocomment - Yes or No with a chance to comment on the answer
• choice-yesnodontknow - Yes or No or Whaaa?
• open - A simple one line input box
• open-textfield - A box for lengthy answers
• choice - A list of choices to choose from
• choice-freeform - A list of choices with a chance to enter something else
• choice-multiple - A list of choices with multiple answers
• choice-multiple-freeform - Multiple Answers with multiple user defined answers
• range - A range of number from which one number can be chosen
• timeperiod - A timeperiod
• custom - custom question using a custom template
• comment - Not a question, but only a comment displayed to the user

*Some of these types, depend on checks or choices. The range question for instance can be controlled by setting the checks to something like "range=1-100 step=1 unit=%". Other questions like the choice-multiple-freeform need a "extracount=10" if ten extra options should be given to the user.

I would love to go into all the details here but time I have not so I my only choice is to kindly refer you to the qprocessor submodule which handles all the question types.*

Next up is the question number. The question number defines the order of questions alphanumerically as long as a number of questions are shown on the same page. The number is also used to refer to the question.

The text of the question is what the user will be asked. There can be one text for each language defined in the settings.py file.

The extra is an additional piece of information shown to the user. As of yet not all questions support this, but most do.

An important aspect of questions (and their parents, QuestionSets) is the checks field. The checks field does a lot of things (possibly too many), the most important of which is to define if a certain question or questionset should be shown to the current subject.

The most important checks on the question are the following:

• required A required question must be answered by the user
• requiredif="number,answer" Means that the question is required if the question with number is equal to answer.
• shownif Same as requiredif, but defining if the question is shown at all.
• maleonly Only shown to male subjects
• femaleonly Only shown to female subjects
• iftag="tag" Question is only shown if the given tag is in the RunInfo

Checks allow for simple boolean expressions like this: iftag="foo or bar" or requiredif="1,yes and 2,no"

Choice

A choice is a possible value for a multiple choice question.

QuestionSet

A bunch of questions together form a questionset. A questionset is ultimately single page of questions. Questions in the same questionset are shown on the same page.

QuestionSets also have checks, with the same options as Questions. There's only one difference, required and requiredif don't do anything.

A questionset which contains no visible questions (as defined by shownif) is skipped.

Answer

Contains the answer to a question. The value of the answer is stored as JSON.

Questionnaire

A questionnaire is a bunch of questionsets together.

Migration of 1.x to 2.0

2.0 added new fields to the questionnaire, but it did so in a backwards compatible way. None of the new fields are mandatory and no changes should be necessary to your existing questionnaire. Since we do not have any relevant testing data however, you might find yourself on your own if it doesn't work. Please file an issue if you think we did something wrong, so we can fix it and help you.

As Django per default does not provide a way to migrate database schemas, we pretty much make use of the bulldozer way of migrating, by exporting the data from one database and import it into a newly created one.

From you existing 1.x site do

python manage.py dumpdata >> export.yaml

Copy your file to your new site and in your new site, create your empty database

python manage.py syncdb

You may then import your data from your old site, which should probably work :)

python manage.py loaddata export.yaml

This of course covers only the data migration. How to migrate your custom tailored site to use questionnaire 2.0 is unfortunately something we cannot really document.

2.0 Postmortem

2.0 was the result of the work we put into seantis questionnaire for our second project with it. We did this project without the help of questionnaire's creator and were pretty much on our own during that time.

Here's what we think we learned:

Seantis.questionnaire is a Framework

More than anything else seantis.questionnaire should be thought of as a framework. Your site has to provide and do certain things for the questionnaire to work. If your site is a customized questionnaire for a company with other needs on the same site you will end up integrating code which will call questionnaire to setup runs and you will probably work through the answer records to provide some sort of summary.

If it was a library you could just work with a nice API, which does not exist.

Don't Go Crazy with Your Checks

We used a fair amount of checks in both questionset and questions to control a complex questionnaire. We offloaded the complexity of the questionnaire into an Excel file defined by the customer and generated checks to copy that complexity into our application.

Though this approach certainly works fine it does not give you a good performance. The problem is, if you have hundreds of questions controlled by runinfo tags, that you end up with most CPU cycles spent on calculating the progress bar on each request. It is precisely for that reason that we implemented the QUESTIONNAIRE_PROGRESS setting (you can learn more about that by looking at the example settings.py).

We managed to keep our rendering time low by doing the progress bar using ajax after a page was rendered. It is only a workaround though. Calculating the progress of a run in a huge questionnaire remains a heavy operation, so for really huge questionnaires one might consider removing the progress bar alltogether. There is still some optimization to be had, but it essentially will remain the slowest part of the questionnaire, because at the end of the day interpreting loads of checks is not something you can do in a fast way, unless your name is PyPy and your programmers are insanly talented.

There are not Enough Tests

There are a couple of tests which do some simple testing, but it's not enough. We failed to add more so far, so we can't shift the blame, but the fact remains still. More tests would mean that more refactoring could be done which would be nice, because there certainly is a need for some refactoring.

The Admin Interface is not Good Enough

Django admin is a nice feature to have, but we either don't leverage it well enough, or it is not the right tool for the questionnaire. In any case, if you are expecting your customer to work with the questionnaire's structure you might have to write your own admin interface. The current one is not good enough.