4. COMPARATIVE EVALUATION OF EACH SOFTWARE PACKAGE
4.5 Questionnaire navigation
For complex household surveys, how interviewers get to questions is just as important as how the questions are asked, if not more so. The defining characteristic of complex household surveys is that they are complex. They contain many unrelated streams of questions. They require several respondents, not all of whom may be available to complete questions in strict sequential order.
35 The quality of navigation controls is judged from the complementary perspectives of questionnaire developers and questionnaire users. From the developer's perspective, navigation is about control. The quality of control is gauged by the degree to which designers can both restrict and strategically relax interviewers' control over navigation, thereby providing the necessary freedom of navigation while imposing some constraints on the questionnaire order. From the interviewer's perspective, navigation is about simplicity. Its simplicity is measured by the ease of use for standard and complex navigation tasks: for answering questions and for correcting any conflicts, particularly across modules, in answers.
Power: Control is the key to power in navigation. Part of control is what the software lets the interviewer do. The ability to move around the questionnaire more freely is a feature that all CAPI packages offer, but that varies vastly in power from one package to another. Some, like CSProX, provide meager tools for non-linear navigation. With a relatively weak toolkit that requires programming, CSProX only offers one of two unsatisfactory options: one the one hand, the default ability to move forward and backwards one questionnaire screen at a time; on the other hand, the marginally
programming -intensive feature of jumping forward to pre-appointed places in the questionnaire, or the considerably more programming-intensive option of allowing specific questions to be skipped at the cost of more program code . Others – like Entryware, Pendragon, and ODK – provide simple tools for
navigation that require little to no programming. These packages offer an in-built ability to jump around a CAPI instrument, but move only to particular questions specified by the interviewer. Still others – like Blaise, CASES, MMIC, and Surveybe – offer more extensive controls over navigation, but require some limited programming. These packages allow the programmer to designate certain sections as "parallel blocks" that can be completed independently of other sections. Blaise, CASES, and MMIC require a small amount of programming to provide this ability. Surveybe provides this by default for all survey sections, yet provides no tools for adjusting this default setting.
An adjunct of what software lets interviewers do is how it lets them do it. CSProX is keyboard- dependent for navigation by default, but can offer some scope for more touch-friendly navigation. Questionnaire movement requires either keyboard shortcuts, selections from the program menu, or clicking a question from the questionnaire case tree. Blaise, Entryware, MMIC, ODK, Pendragon offer touch-friendly buttons to move from one survey screen to the next. Blaise and Surveybe offer a panel of navigation tabs to move from one module to another. Blaise and CASES offer programmable jump menus that enable interviewers to access only those sections that are currently relevant and allowed. The other parts of power are what software prevents interviewers from doing, and how it constrains them. By default, CSProX constrains interviewers to the programmed path of the interview, not as an option but as a rigid default, which does not provide any navigational power whatsoever. Entryware, Pendragon, and ODK err in the opposite direction by allowing relatively free movement that is
uncontrolled by any constraints, which gives rise to its own set of problems. By careful programming, CSProX provides comparable freedom, but unlike the above does so at the cost of developing more program code. However, Blaise, CASES, MMIC, and Surveybe provide navigation with some constraints. CASES and MMIC allow developers to provide interviewers with non-linear navigation abilities, but also endow programmers with tools for constraining the scope of non-linearity. CASES and MMIC can prevent interviewers from entering certain blocks of a questionnaire without having first completed other blocks. MMIC, for example, can conditionally enable (and disable) entire survey modules based on whether or not a prior module has been completed. Blaise and Surveybe provide much freer movement than either CASES or MMIC, but also have lesser ability to constrain movement. Their only constraints are the normal ones: preventing interviewers from moving down an irrelevant branch of
36 questioning. Yet the example of completing a roster in a questionnaire demonstrates that this is still a feat. In Surveybe the enumerator sees an entire roster at once, can enter data on any person in that roster, and move forward into any sub-subsection of the roster for any given person. Thanks to
programming tools, Surveybe can prevent completely free movement by disabling the sub-sections of a roster until such time as an interviewer has provided requisite information on whether that sub-section is relevant for a given person.
CASES and MMIC offer the most power in what navigation they allow and constrain interviews to have, although this power comes at programming costs. Both packages offer non-linear completion through designating blocks of a questionnaire as "parallel blocks". Both packages can prevent interviewers from completely non-linear movement by requiring that certain modules be completed before others can be begun.
Blaise and Surveybe offer non-linear completion with much greater ease than CASES and MMIC. Blaise requires a small amount of programming to designate sections as "parallel blocks" and then creates tabs for interviewers to access them. Surveybe offers this feature with no programming requirements at all. Questionnaire sections in Surveybe are by default "parallel blocks". However, these packages fail to constrain non-linear navigation significantly.
Simplicity: Powerful controls, to be useful, must also be simple. Most software packages offer buttons for forward and backward movement, but a few also offer many more user-friendly navigation features. Blaise and Surveybe present a clickable navigation panel for moving from one independent module to another. CASES provides programmable buttons that offer users a more sophisticated set of navigation options. MMIC can create color-coded buttons that indicate which modules an interviewer has
completed and where they are allowed to go next. These all result in the simplicity of one-click navigation.
However, Surveybe leads the pack for simplicity of navigation for interviewers. Through its navigation panel, it allows interviewers to move freely from one module to another by a click of a button at any point during the interview. The only other CAPI packages in which this readily appears are Blaise and Entryware. Through its unique questionnaire layout where buttons provide easy access to more detailed questions, Surveybe renders the otherwise confusing or tiresome navigation through forward and backwards buttons unnecessary. Through all of its navigation features, Surveybe provides interviewers a common mode of moving around: click on the labeled item where you want to go. This holds true for sections, each of which has its own labeled tab, and for sub-sections of follow-up questions, which also have a button with a brief description of their content.
Another aspect of simplicity in navigation concerns the ease with which an interviewer can move around to fix any errors or inconsistencies that the CAPI package may have identified. Although all software under consideration offer the traditional means of paging forwards and backwards to fix an errant field, only four of them offer an interactive means of doing so that exploits non-linear navigation, in two fundamentally different ways. Blaise and MMIC both prompt interviewers to revise potentially errant fields and provide them with the on-screen means for skipping to those fields. CSProX offers
comparable functionality through a pop-up menu. This is all triggered when errors occur, and can be designed as "soft" or "hard" errors – that is, as possible errors that may be bypassed or as definite errors that must be addressed immediately. Surveybe takes a slightly different tack. It offers the interviewer the means for jumping directly to problematic or potentially problematic fields with a single click on the appropriate row displayed in the list of errors. It differs in its approach, however, in that these errors
37 may be raised and addressed at any point, after the interviewer has elected to "validate" part or all of a questionnaire. This allows enumerators to continue their work and make revisions later, while the other packages force errors to be addressed immediately. These are differences in design philosophy, but the overall approach –interactive error correction and provision of links to problem fields – is the same. Surveybe is both powerful for developers and simple for interviewers. Much of Surveybe's navigation capabilities come from design features that require no programming. Other packages, like MMIC, come close to challenging Surveybe's power, but replicating Surveybe's features with such functionality comes at the cost of program code.