Randomize Choices and Question Order in XLSForm

To shuffle the choices of a question in XLSForm, write randomize=true in the parameters column of a select_one or select_multiple row; add seed=${my_seed} to keep one order per interview. There is no column that shuffles the order of questions, but a repeat with randomize() and selected-at() does it, and once(random()) assigns a random form version. Every example below compiles with pyxform and ODK Validate.

Download example form (.xlsx)All five examples in one XLSForm, one group each. Tested with pyxform and ODK Validate; free to reuse under CC BY 4.0.Download

How do I randomize the order of choices?

Respondents pick options near the top of a list more often, especially when the enumerator reads the list aloud. Shuffling the list spreads that bias across all options. A food security survey asking where the household buys staples:

type	name	label	parameters
select_one source	staple_source	Where does the household buy most of its rice?	randomize=true
select_multiple source	other_sources	Where else did the household buy food in the last 7 days?	randomize=true

◀ ▶surveychoicessettings+

list_name	name	label
source	1	Local market
source	2	Village shop
source	3	Supermarket
source	4	Mobile vendor
source	5	Cooperative or fair price shop

◀ ▶surveychoicessettings+

Only the display order changes. The data still stores 1 for Local market, whatever position it appeared in, so analysis code needs no change.

Without a seed, the order changes every time the question is shown, including when the enumerator goes back to it or reopens a saved draft. That is confusing in the field and makes it impossible to know which order a respondent saw.

How do I keep the same order for the whole interview?

Give the shuffle a seed. A seed that is fixed for the submission gives one order per interview, and the same order every time that interview is reopened. ODK's own documentation uses the fractional part of the current time:

type	name	label	calculation	parameters
calculate	order_seed		once(substr(decimal-date-time(now()), 8))
select_one brand	soap_brand	Which brand of soap does the household use most?		randomize=true seed=${order_seed}
calculate	soap_order		once(join(' ', randomize(instance('brand')/root/item/name, ${order_seed})))

◀ ▶surveychoicessettings+

list_name	name	label
brand	1	Lifebuoy
brand	2	Lux
brand	3	Keya
brand	4	Meril

◀ ▶surveychoicessettings+

• once() is what holds the seed. Without it, the time is read again on later evaluations and the order can change. The unique ID guide explains how once() behaves.
• seed needs randomize=true in the same cell. The two are separated by a space.
• soap_order saves the shuffled codes, for example 3 1 4 2, using the same function and seed as the question. Check a few test submissions against what the phone showed before relying on it for an order-effects analysis.

Any number unique to the interview works as a seed: a respondent ID, or the last digits of a phone number.

If the list needs a "Don't know", ask about it first ("Does the household have a preferred brand?") and show the randomized list only to those who say yes. For an "Other" option with a text box, see Other, please specify in XLSForm.

How do I randomize the order of questions or sections?

XLSForm has no setting for question order. The method that scales, described on the ODK forum, is to shuffle the block numbers once, then put the blocks in a repeat that runs once per block, and show only one block in each round. Here a WASH survey asks its water, sanitation and hygiene sections in a random order:

type	name	label	calculation	relevant	repeat_count
calculate	block_order		once(join(' ', randomize(instance('wash_block')/root/item/name)))
begin_repeat	section	WASH section			3
calculate	block_id		selected-at(${block_order}, position(..) - 1)
begin_group	water	Drinking water		${block_id} = '1'
select_one yes_no	water_treated	Did the household treat its drinking water in the last 7 days?
end_group	water
begin_group	sanitation	Sanitation		${block_id} = '2'
select_one yes_no	latrine_shared	Does the household share its latrine with other households?
end_group	sanitation
begin_group	hygiene	Handwashing		${block_id} = '3'
select_one yes_no	soap_seen	Is soap present at the handwashing place?
end_group	hygiene
end_repeat	section

◀ ▶surveychoicessettings+

list_name	name	label
wash_block	1	Drinking water
wash_block	2	Sanitation
wash_block	3	Handwashing

◀ ▶surveychoicessettings+

list_name	name	label
yes_no	1	Yes
yes_no	0	No

◀ ▶surveychoicessettings+

How it works:

• block_order reads the codes of the wash_block list, shuffles them with randomize() and joins them into text such as 2 3 1. No question uses the list; it only feeds the shuffle.
• The repeat runs 3 times. In each round, selected-at() takes the next code from block_order (it counts from 0, hence position(..) - 1), and only the group with that code is relevant.
• block_order is saved with the submission, so you know which order each respondent got.

The cost is in the data: the answers land on the repeat sheet, one row per section, with blanks for the groups that were not shown in that round. Merge them back to one row per household at analysis time. For more on how repeats are exported, see XLSForm repeat groups.

To shuffle questions inside one section, use the same pattern with one question per round instead of one group.

How do I show a random version of a question (split-ballot)?

Survey experiments often show half the respondents one wording and half another. Draw a random number once and use it in relevant:

type	name	label	calculation	relevant
calculate	arm		once(int(random() * 2) + 1)
begin_group	version_a	Version A		${arm} = 1
integer	days_ill_a	In the last 30 days, how many days was the child ill?
end_group	version_a
begin_group	version_b	Version B		${arm} = 2
integer	days_ill_b	In the last 30 days, how many days did the child have fever, diarrhoea or cough?
end_group	version_b

◀ ▶surveychoicessettings+

int(random() * 2) + 1 gives 1 or 2 with equal chance; change the 2 for more arms. once() matters again: random() can be evaluated more than once during an interview, and without once() a respondent could switch arm partway through. The arm value is in the data, so you can compare the two versions. For the logic in relevant, see the XLSForm skip logic guide.

How do I randomize the order of a few sections without a repeat?

With 2 or 3 sections, you can avoid the repeat: list each order as its own group and show one at random. Two sections have two orders:

type	name	label	calculation	relevant
calculate	path		once(int(random() * 2) + 1)
begin_group	path_1	Income then expenses		${path} = 1
integer	income_1	What was the household income last month?
integer	spend_1	What did the household spend on food last month?
end_group	path_1
begin_group	path_2	Expenses then income		${path} = 2
integer	spend_2	What did the household spend on food last month?
integer	income_2	What was the household income last month?
end_group	path_2
calculate	income		coalesce(${income_1}, ${income_2})
calculate	food_spend		coalesce(${spend_1}, ${spend_2})

◀ ▶surveychoicessettings+

Each question needs a different name in each path, so coalesce() joins the copies back into one income and one food_spend column. Three sections have 6 orders and four have 24, so past three sections the repeat method above is less work.

Common randomization mistakes and how to fix them

These are the messages pyxform and ODK Validate return, which is what ODK Central, KoboToolbox or the XLSForm converter shows when it rejects the form:

You wrote	Write instead	Error you will see
randomize=yes	randomize=true	randomize must be set to true or false: 'yes' is an invalid value
randomise=true	randomize=true (with a z)	Accepted parameters are 'randomize, seed'. The following are invalid parameter(s): 'randomise'.
seed=${order_seed} alone	randomize=true seed=${order_seed}	Parameters must include randomize=true to use a seed.
randomize=true on a text question	only on select_one or select_multiple	Accepted parameters are 'rows'. The following are invalid parameter(s): 'randomize'.
seed=${orderseed} (misspelt name)	the exact name of the seed calculate	Could not find the name 'orderseed'.
randomize(instance('brand')/root/item/name) in a calculate	wrap it: once(join(' ', randomize(...)))	XPath evaluation: type mismatch
randomize=true in the appearance column	move it to parameters	No error, but the choices are not shuffled
int(random() * 2) + 1 without once()	once(int(random() * 2) + 1)	No error, but the version can change during the interview

For other messages, see common XLSForm errors or paste the form into the free XLSForm validator.

Does this work the same in ODK, KoboToolbox and SurveyCTO?

ODK and KoboToolbox use the same syntax. SurveyCTO randomizes choices differently:

Platform	Shuffle choices	Keep one order per interview	Keep some choices fixed
ODK (XLSForm)	randomize=true in parameters	add seed=${field}	Not possible
KoboToolbox (XLSForm upload)	Same as ODK	Same as ODK	Not possible
SurveyCTO	randomized in the appearance column, or "Automatically randomize choice order" in the form designer	See SurveyCTO's documentation	randomized(0, 1) keeps the top 0 and bottom 1 choices in place

The question order and split-ballot methods use random(), once(), randomize() and selected-at(), which are ODK XPath functions. They compile for ODK Collect and KoboToolbox; preview the form in Enketo (Kobo's web form) before fieldwork. SurveyCTO has its own guidance on randomizing section order, so check it before reusing the repeat method there.

When randomizing is worth it

Randomize unordered lists that respondents hear read aloud or see on a tablet: brands, sources, reasons, preferences. Leave alone lists with a natural order (age bands, frequency scales, Likert items) and lists the enumerator needs in a fixed place to find quickly, such as districts. Shuffling a long list of districts only slows the interview down; for those, a searchable list works better (see the XLSForm appearance column).

Build my first form free

Next: XLSForm appearance column →