Skip to contents

Indicators of automated or careless responding for one participant

dq_flags.Rd

Sums the page records of one participant (see dq_parse()) and reports, per indicator, what was observed and whether it crosses the threshold at which it is worth a look.

Usage

dq_flags(
  d,
  probe = NA,
  srv_ua = NA,
  sig_agent = NA,
  ch_platform = NA,
  ch_mobile = NA
)

Arguments

d

a data frame of page records, as returned by dq_parse().

probe

what was stored in the participant's agent_probe items, as one string (people never see these fields, so any text is flagged).

srv_ua

the User-Agent header stored by a browser item. It is compared with the User-Agent the page read.

sig_agent

the Signature-Agent header stored by a server HTTP_SIGNATURE_AGENT item. AI agents that identify themselves send it.

ch_platform, ch_mobile

the client hints stored by server HTTP_SEC_CH_UA_PLATFORM and server HTTP_SEC_CH_UA_MOBILE items. They are compared with srv_ua: a User-Agent rewritten on its way to the server leaves them as they were.

Value

A data frame with one row per indicator and the columns group ("environment": the browser and the request, "input": how clicks and text were made, "attention": time, leaving the page, copy and paste), indicator, value (what was observed, as text) and flag (TRUE when the indicator crosses its threshold; purely informative indicators are never flagged). If d has no record, a single unflagged row says so.

Details

No single indicator proves anything. Several input indicators together, or any text in an agent_probe field, are strong signs that a program made the input; the attention indicators (leaving the page, copying and pasting) are context about how a person answered. Review flagged participants by hand before excluding anyone.

See also

dq_flags_by_session() for a whole results table.

Examples

records <- c(
  '{"v":2,"ld":1,"t":41000,"keys":0,"txt_ch":80,"fld_nokey":1,"pd":4,"pd_ctr":4}',
  '{"v":2,"ld":1,"t":9000,"keys":0,"txt_ch":0,"pd":2,"pd_ctr":2}'
)
flags <- dq_flags(dq_parse(records), probe = "yes, a language model")
flags[flags$flag, ]
#>          group                                             indicator
#> 13 environment                              agent_probe field filled
#> 25       input      text answers mostly not typed or pasted (fields)
#> 27       input share of text typed via keys (keys / chars in fields)
#> 37       input                       mouse presses / at exact centre
#>                    value flag
#> 13 yes, a language model TRUE
#> 25                     1 TRUE
#> 27                     0 TRUE
#> 37                 6 / 6 TRUE