S anda ds o anspa en AI in Human
Resou ce Managemen (TRANKI) – Resea ch
da a
Ka ha ina Simbeck*Yannick Kal †
HTW Be lin Uni e si y o Applied Sciences
Table o con en s
Disclaime 2
De ailson heinqui y............................................ 2
De ailson heda a ............................................. 3
Using he da a-se 3
Folde s uc u eand ile ee........................................ 3
Reading CSV, a aching a iable names and labels . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Inspec ing heda a-se ........................................... 4
Codebook 4
I ems and indices 5
TechnologyAccep anceModel ...................................... 5
AILi e acy.................................................. 6
Technophobia................................................ 7
Appendix 7
Dashboa dsExpe imen 1......................................... 7
Dashboa dsExpe imen 2......................................... 8
Dashboa dsExpe imen 3......................................... 9
Con ac de ails 10
License 11
*En elope[email p o ec ed] �h ps://o cid.o g/0000-0001-6792-461X
†En elope[email p o ec ed] �h ps://o cid.o g/0000-0003-1595-175X
1
Disclaime
The ollowing da a-se was collec ed in Feb ua y 2025 o he esea ch P ojec “TRANKI – S anda ds
o anspa en AI”, unded by he Hans-Böckle -Founda ion
1
. The goal was o assess he e ec s o AI
li e acy on he in e p e a ion o AI-enhanced use in e aces o HR so wa e, and how explainable AI (XAI)
elemen s a ec in e p e a ion esul s. The p ojec ’s esea ch ques ions we e:
•
Wha app oaches a he use in e ace le el help make AI sys ems in human esou ces anspa en o
use s?
•
Wha di e ences in anspa ency equi emen s and pe cep ions exis be ween employees, manage s,
and HR expe s?
All i ems a e included wi h hei Ge man labels and English ansla ions. The ansla ion p ocess was
au oma ed, using LLM model
mis al-medium-2508
, and p oo - ead by he au ho s. Please eel ee o
ansla e u he as needed.
The expe imen s p esen ed he pa icipan s wi h mock-up sc eens o HR-so wa e ools. The basis we e
eal-li e exis ing ools ha ad e ised hei AI componen s. In a i s s ep, we esea ched ool endo s and
sc eened hei homepages o sc eensho s and demons a o s. Then, we ec ea ed he use in e aces wi h
sligh de i a ions in wo e sions: a) baseline, b) wi h added XAI elemen s o explain he AI esul s. We
used Moqups.2The dashboa d sc eensho s a e in he appendix.
The expe imen s conduc ed each include an objec i e assessmen o he co ec o inco ec in e p e a ion
o in o ma ion om an AI-suppo ed dashboa d wi h/wi hou explainable AI elemen s. This da a-se
con ains he ans o med and e alua ed i ems and accumula ed poin s o co ec answe s. (Va iables:
Exp1_objsco e_pos
,
Exp1_objsco e_p e
,
Exp2_objsco e_pos
,
Exp2_objsco e_p e
,
Exp3_objsco e_pos
,
Exp3_objsco e_p e
– whe e
_pos
indica es sco es o XAI-enhanced dashboa ds, and
_p e
anilla
dashboa ds.)
De ails on he inqui y
•Design: Expe imen al design, su ey, c oss-sec ional s udy
•P e es : 15 pa icipan s in Decembe 2024.
•Sample:
–Size: 𝑁 = 427
–Rec ui men : ISO 20252:2019-ce i ied sampling p o ide
–
Inclusion c i e ia: cu en ly employed o employed wi hin he las 12 mon hs in he ield o HR
(any le el)
•Su ey pe iod: 2025-02-01 – 2025-02-28
•E hics: In o med consen om pa icipan s be o e inqui y
1
URL: h ps://www.boeckle .de/de/suche gebnis- o schungs oe de ungsp ojek e-de ailsei e-2732.h m?p ojek =2022-797-2.
(G an No. 2022-797-2).
2We hank Thao Do o assis ing us wi h he esea ch and c ea ion p ocess o he expe imen s’ dashboa ds.
2
De ails on he da a
We emo ed 17 implausible cases, e aining
𝑁 = 410
cases o analysis. Implausible cases demons a ed
un ealis ic ela ionships be ween age and wo k expe ience, o example: age 50 wi h 45 yea s o expe ience
in HR, o ou lie s in epo ed income.
We excluded pa icipan s whe e
𝑎𝑔𝑒 − 𝑤𝑜𝑟𝑘𝑒𝑥𝑝𝑒𝑟𝑖𝑒𝑛𝑐𝑒
was below 16 yea s, meaning any pe son who would
ha e been younge han 16 yea s when hey began wo king was excluded. Addi ionally, we emo ed income
ou lie s below
300
€ and abo e
10000
€ pe mon h. The p o ided da a-se has been il e ed acco dingly and
includes hese 410 cases (c . he ollowing codebox).
exclude <- d _o %>%
il e (((2025 -Q6_4) -Q4_4) <16 |Q5_4 <300 |Q5_4 >10000)%>%
pull(ID)
d _o <- d _o %>%
il e (!ID %in% exclude)
The expe imen al g oup assignmen s o expe imen s 2 and 3 a e:
• IF: Impo an ea u es (i.e., Fea u e Impo ance)
• CF: Wha i (i.e., Coun e ac uals)
• MC: Model c i e ia
Using he da a-se
Folde s uc u e and ile ee
FOLDER-OPENT anki-da a/
│
├── FILE-ALT anki_da a.cs # CSV Da a se
├── FILE-ALT a names.cs # Va iable names Ge man
├── FILE-ALT a labels.cs # Va iable labels Ge man
├── FILE-ALT a names_engl.cs # Va iable names English
├── FILE-ALT a labels_engl.cs # Va iable labels English
│
├── FILE-ALTLICENSE.md # License
├── FILE-ALTREADME.pd # This ile
└── FILE-ALTCHANGELOG.md # Ve sion his o y and changes
Reading CSV, a aching a iable names and labels
Run he ollowing
R
-code i s o load he da a and assign a iable names and labels. O he wise, use he
p o ided SPSS ile.
3
lib a y(labelled)
da a <- ead.cs ("da a/ anki_da a.cs ")
# Please selec : ei he
a names <- ead.cs ("da a/ a names.cs ")# o o iginal Ge man a iable names
a labels <- ead.cs ("da a/ a labels.cs ")# o o iginal Ge man a iable labels
# o
# a names <- ead.cs ("da a/ a names_engl.cs ") # o ansla ed English a iable names
# a labels <- ead.cs ("da a/ a labels_engl.cs ") # o ansla ed English a iable labels
# Add a names
o (i in 1:n ow( a names)) {
i (!is.na( a names$ a iable_label[i])) {
a _name <- as.cha ac e ( a names$ a iable_name[i])
i ( a _name %in% names(da a)) {
a (da a[[ a _name]], "label") <- a names$ a iable_label[i]
}
}
}
# Add labels
o ( a in unique( a labels$ a iable)) {
i ( a %in% names(da a)) {
a _labels <- a labels[ a labels$ a iable == a , ]
labels <- se Names( a _labels$ alue, a _labels$label)
a _label <- a (da a[[ a ]], "label")
da a[[ a ]] <- se _ alue_labels(da a[[ a ]], labels)
i (!is.null( a _label) && is.null(a (da a[[ a ]], "label"))) {
a (da a[[ a ]], "label") <- a _label
}
}
}
Inspec ing he da a-se
Some example que ies.
p in (da a$Q1_1)
summa y(da a$Q6_4)
plo (da a$Q4_4, da a$Q5_4)
Codebook
To c ea e a codebook, use he ollowing syn ax.
4
kni ::op s_chunk$se (
wa ning = TRUE,
message = TRUE,
e o = TRUE,
echo = TRUE
)
ggplo 2:: heme_se (ggplo 2:: heme_bw())
pande ::pande Op ions(" able.spli . able",In )
lib a y(codebook)
lib a y(ggplo 2)
lib a y(summa y ools)
codebook <- d Summa y(da a,
a numbe s = FALSE,
na.col = FALSE,
s yle = "mul iline",
plain.ascii = FALSE,
headings = TRUE,
max.dis inc . alues = 5,
mp.img.di = "./ mp")
# Rende he Codebook
p in (codebook, me hod = " ende ")
I ems and indices
The ques ionnai e inco po a es se e al hi d-pa y indices, pa icula ly o assessing cons uc s ela ed o he
Technology Accep ance Model (TAM),AI li e acy, and echnophobia. The sec ions below e alua e
he psychome ic quali y o hese measu es and p o ide e e ences o he espec i e li e a u e.
Technology Accep ance Model
Technology Accep ance Model ollows he concep o Da ies 1989
3
bu wi h indi idually gene a ed i ems
o he ou sub-indices a i ude owa ds use (ATU),pe cei ed use ulness (PU),pe cei ed ease o use (PEOU), and
ex e nal ac o s (EXT).
Pa icipan s we e p esen ed wi h he espec i e i ems i hey me ei he o he ollowing c i e ia: 1. Thei
employe explici ly pe mi ed he use o AI ools in wo k p ocesses, o 2. They independen ly u ilized AI
ools in hei wo k— ega dless o employe consen .
3
Da is, F. D. (1989). Pe cei ed Use ulness, Pe cei ed Ease o Use, and Use Accep ance o In o ma ion Technology. MIS
Qua e ly, 13(3), 319. h ps://doi.o g/10.2307/249008
5
Table 1: O e iew o he scale TAM
Scale No. I ems C onbach’s 𝛼Mean (SD) 𝑚𝑒𝑎𝑛 (𝑟𝑖𝑡.𝑐𝑜𝑟𝑟 )
ATU 2 (3) 0.73 3.73 (1.25)0.58
PU 3 0.80 3.79 (0.96)0.64
PEOU 3 0.77 3.81 (0.89)0.60
EXT 3 0.70 3.48 (0.96)0.53
The i em
Q4_2 9
has o be in e ed. Howe e , he i em s a is ics sugges ha esponses we e ambiguous.
The e o e, we decided o exclude i om he ATU sub-scale and u he analysis.
AI Li e acy
AI li e acy was measu ed using he Scale o he Assessmen o Non-Expe s’ AI Li e acy (SNAIL), de eloped
by Laupichle e al.
4
The ques ionnai e i ems we e d awn om he Ge man ansla ion p o ided in he
appendix o Laupichle e al.5The o iginal scale comp ises 30 i ems ac oss h ee dimensions:
•Technical unde s anding (TU)
•C i ical app aisal (CA)
•P ac ical applica ion (PA)
Fo his s udy, 15 i ems wi h he highes ele ance o human esou ce (HR) managemen we e selec ed o
ensu e con ex ual app op ia eness.
The index showed excep ional quali y wi h high C onbach’s 𝛼in all h ee sub-dimensions.
Table 2: O e iew o he scale AI Li e acy
Scale No. I ems C onbach’s 𝛼Mean (SD) 𝑚𝑒𝑎𝑛 (𝑟𝑖𝑡.𝑐𝑜𝑟𝑟 )
TU 5 0.92 2.85 (1.28)0.79
CA 5 0.90 3.29 (1.22)0.75
PA 5 0.89 3.10 (1.24)0.73
Fu he discussions o he quali y o he index can be ound in ou publica ions.6
4
Laupichle , M. C., As e , A., Ha e kamp, N., & Raupach, T. (2023). De elopmen o he “Scale o he assessmen o non-expe s’
AI li e acy” – An explo a o y ac o analysis. Compu e s in Human Beha io Repo s, 12, 100338. h ps://doi.o g/10.1016/j.chb .2
023.100338
5Laupichle , M. C., As e , A., Pe schewski, J.-O., & Schleiss, J. (2023). E alua ing AI Cou ses: A Valid and Reliable Ins umen
o Assessing A i icial-In elligence Lea ning h ough Compa a i e Sel -Assessmen . Educa ion Sciences, 13(10), 978. h ps:
//doi.o g/10.3390/educsci13100978
6
Kal , Y., & Simbeck, K. (2025). Explained, ye misunde s ood: How AI Li e acy shapes HR Manage s’ in e p e a ion o
Use In e aces in Rec ui ing Recommende Sys ems. In M. Kaya, T. Boge s, G. Bied, C. Johnson, & J.-J. Deco e (H sg.),
P oceedings o he 5 h Wo kshop on Recommende Sys ems o Human Resou ces (RecSys in HR 2025) (Ve sion 1, Bd. 4046). CEUR.
h ps://ceu -ws.o g/Vol-4046/RecSysHR2025-pape _3.pd
6
Technophobia
The scale was o iginally de eloped by Sinko ics
7
o assess esis ance owa d eme ging au oma ed echnologies,
wi h a ocus on he hen-no el ATM machines. Fo his s udy, he scale was adap ed o e alua e employees’
nega i e a ec and app ehensions ega ding he inc easing in eg a ion o AI in HR wo kplace con ex s.
F om he o iginal 13 Ge man-language i ems, six we e selec ed based on hei ele ance o AI-speci ic
anxie ies in HR se ings. Two i ems (
Q2_3 4
and
Q2_3 5
) equi e e e se sco ing p io o analysis. The
adap ed scale demons a ed adequa e in e nal consis ency (C onbach’s
𝛼 = .72
), wi h a mean sco e o
𝑀 = 3.07 (𝑆𝐷 = 0.74).
Appendix
Dashboa ds Expe imen 1
Figu e 1: Baseline
Figu e 2: Va ia ion 1: Impo an Fea u es
7
Sinko ics, R. R. (2003). Technophobie. Zusammens ellung sozialwissenscha liche I ems und Skalen (ZIS).h ps://doi.o g/10.6102/ZI
S62
7
Dashboa ds Expe imen 2
Figu e 3: Baseline
Figu e 4: Va ia ion 1: Impo an Fea u es
Figu e 5: Va ia ion 2: Coun e Fac uals
Figu e 6: Va ia ion 3: Model c i e ia
8
Dashboa ds Expe imen 3
Figu e 7: Baseline
9
