PEPS-Y Political Discussion: Private Talk, Workplace Talk, and the Talk Gap

1 Setup

This report follows preregs/political_discussion_prereg and the theory draft in papers/privateworkplace political-talk gap_theory_and_hyps_20260223.docx. The pilot file contains the core discussion items (poldisfam, poldiscol), political interest, institutional trust, public-service evaluations, a usable political-opinion minority indicator, and a cross-cutting-friendship item (frndpol).

The cleaned pilot sample contains 355 interviews. Of these, 292 report having work colleagues, and 292 contribute to the within-person talk-gap analysis. The mean private-talk score is 4.39, the mean workplace-talk score among respondents with colleagues is 3.67, and the average private-minus-workplace gap is 0.73.

2 Preregistered Hypotheses

hypothesis status proxy
H1 Political interest Tested now Political interest across private talk, workplace talk, and the gap
H2 Perceived minority positioning Tested now Political-opinion minority (idminor7) as the main prereg proxy
H3 Cross-cutting friendships Tested now Friends with different political views (frndpol)
H4 Institutional trust Tested now Institutional trust index
H5 Public service experiences Tested now Reversed service-quality index so higher values mean more negative experiences
H7 Gender × migration-status gap Tested now Gender, migrant background, and their interaction in workplace and gap models
  • H1 Political interest. More interested respondents should talk about politics more often in both settings, especially in private settings.
  • H2 Perceived minority positioning. People who see their political views as being in the minority should talk less at work and show a larger private-work gap.
  • H3 Cross-cutting friendships. Having friends with different political views should increase private talk and reduce workplace self-suppression.
  • H4 Institutional trust. Higher trust should make workplace talk easier and narrow the talk gap.
  • H5 Public service experiences. More negative public-service experiences should feed private complaint talk and widen the gap.
  • H7 Gender × migration status. Women with a migrant background should show a smaller gap because workplace talk is relatively more common for them.

3 Testing Strategy

The core design follows the preregistration closely. The two talk questions use the same 7-point frequency scale running from 1 = Never to 7 = Every day. One asks how often respondents discuss national politics with family and friends, and the other asks how often they discuss it with work colleagues. I estimate ordered-logit models for these two original ordinal outcomes. I then build a simple within-person gap measure defined as private talk - workplace talk. So if a respondent answers 5 for private talk and 3 for workplace talk, the resulting gap is 2, which means politics is discussed more often in private life than at work. A gap of 0 means the two settings are equally frequent, and negative values mean the respondent talks about politics more often at work than in private settings. All models adjust for age, education, employment status, left-right placement, and country fixed effects. The workplace and gap models are restricted to respondents who report having work colleagues.

I treat idminor7 as the main prereg proxy for perceived minority positioning because it is the most direct match to political expression risk. For public-service experiences, I reverse the service-quality index so that larger values mean more negative experiences; this keeps the coefficient direction aligned with the theory text. I also run the preregistered OLS functional-form checks for the two talk outcomes, the minority-positioning moderation tests, and the optional four-category gap typology when the multinomial model is available.

4 Descriptive Checks

The pilot reproduces the basic pattern that motivates the paper: respondents report talking about national politics more often with family and friends than with colleagues.

The next figures show the same three outcomes in a more compact summary form: mean private-talk frequency, mean workplace-talk frequency, and the mean talk gap. The first chart shows overall averages. The next three switch the comparison around: each outcome appears on the x-axis, and the colored bars compare gender, migrant background, or ideology within that outcome. That makes it easier to see, for example, whether women and men differ more in private talk, workplace talk, or the size of the gap itself.

5 Main Models

The tables below compare the same five mechanisms across the three main outcomes. Positive coefficients in the gap model mean a wider tendency to keep politics in private life rather than discuss it at work.

5.1 Private Talk

term_label estimate conf_int p_value
Political interest 1.334*** [1.074, 1.603] <0.001
Political-opinion minority 0.595 [-0.399, 1.607] 0.242
Public-service dissatisfaction 0.489** [0.191, 0.791] 0.001
Institutional trust 0.240 [-0.071, 0.552] 0.131
Cross-cutting friendships -0.118 [-0.346, 0.110] 0.312

5.2 Workplace Talk

term_label estimate conf_int p_value
Political interest 0.826*** [0.572, 1.088] <0.001
Public-service dissatisfaction 0.227 [-0.079, 0.535] 0.147
Institutional trust 0.039 [-0.289, 0.368] 0.815
Cross-cutting friendships -0.170 [-0.413, 0.073] 0.170
Political-opinion minority -0.522 [-1.600, 0.530] 0.333

5.3 Talk Gap

term_label estimate conf_int p_value
Political-opinion minority 0.738 [-0.401, 1.876] 0.205
Political interest 0.297* [0.069, 0.525] 0.011
Public-service dissatisfaction 0.127 [-0.164, 0.418] 0.393
Institutional trust 0.091 [-0.217, 0.400] 0.562
Cross-cutting friendships -0.010 [-0.221, 0.202] 0.929

5.4 Classic Regression Tables

The compact table above is useful for quick comparison across the three outcomes. The tables below present the same models in a more conventional regression format.

Characteristic

Private talk

Workplace talk

Talk gap

log(OR)

1

95% CI

1

log(OR)

1

95% CI

1

Beta

95% CI

1

p-value
Political interest 1.33 1.07, 1.60 0.826 0.572, 1.09 0.297 0.088, 0.505 0.005
Political-opinion minority 0.595 -0.399, 1.61 -0.522 -1.60, 0.530 0.738 -0.195, 1.67 0.12
Cross-cutting friendships -0.118 -0.346, 0.110 -0.170 -0.413, 0.073 -0.010 -0.226, 0.206 >0.9
Institutional trust 0.240 -0.071, 0.552 0.039 -0.289, 0.368 0.091 -0.204, 0.387 0.5
Public-service dissatisfaction 0.489 0.191, 0.791 0.227 -0.079, 0.535 0.127 -0.142, 0.396 0.4
1

OR = Odds Ratio, CI = Confidence Interval

The strongest pilot result is H1: political interest predicts more talk in both settings, and it also widens the private-minus-workplace gap because the private-talk association is clearly stronger. Beyond that, the evidence for the social-risk mechanisms is much thinner. Political-opinion minority status points in the expected direction, but the estimates are imprecise. Cross-cutting friendships do not show the predicted main-effect pattern. Institutional trust is also weak in the main models. The clearest non-interest signal comes from public-service dissatisfaction: respondents with more negative service evaluations talk more in private settings, which fits the idea of private complaint talk, but the gap effect is much less precise.

6 Hypothesis Assessment

hypothesis hypothesis_text how_tested pilot_verdict evidence
H1 Political interest increases private talk and workplace talk, and the increase is stronger in private settings. Main ordered models for private and workplace talk plus the OLS gap model. Confirmed in the pilot Private talk: beta = 1.334, 95% CI [1.074, 1.603], p = 0.000; workplace talk: beta = 0.826, 95% CI [0.572, 1.088], p = 0.000; talk gap: beta = 0.297, 95% CI [0.069, 0.525], p = 0.011.
H2 Perceiving oneself as a political-opinion minority reduces workplace talk and widens the private-workplace gap. Main workplace-talk and gap models using `idminor7` as the preregistered minority proxy. Mixed pilot evidence Workplace talk: beta = -0.522, 95% CI [-1.6, 0.53], p = 0.333; talk gap: beta = 0.738, 95% CI [-0.401, 1.876], p = 0.205.
H3 Cross-cutting friendships increase private talk and reduce the private-workplace gap. Main private-talk and gap models using the cross-cutting-friendship measure. Not confirmed in the pilot Private talk: beta = -0.118, 95% CI [-0.346, 0.11], p = 0.312; talk gap: beta = -0.01, 95% CI [-0.221, 0.202], p = 0.929.
H4 Higher institutional trust increases workplace talk and reduces the gap. Main workplace-talk and gap models using the institutional-trust index. Not confirmed in the pilot Workplace talk: beta = 0.039, 95% CI [-0.289, 0.368], p = 0.815; talk gap: beta = 0.091, 95% CI [-0.217, 0.4], p = 0.562.
H5 More negative public-service experiences increase private talk and widen the gap. Main private-talk and gap models using the reversed service-quality measure, so higher values mean more negative experiences. Mixed pilot evidence Private talk: beta = 0.489, 95% CI [0.191, 0.791], p = 0.001; talk gap: beta = 0.127, 95% CI [-0.164, 0.418], p = 0.393.
H7 Women with a migrant background show a smaller talk gap because they are relatively more likely to discuss politics at work. Separate workplace-talk and gap models with gender, migrant background, and their interaction. Mixed pilot evidence Workplace interaction: beta = 0.514, 95% CI [-1.077, 2.109], p = 0.525; gap interaction: beta = -0.942, 95% CI [-2.027, 0.142], p = 0.090.

The pilot therefore supports a narrower story than the theory section initially suggests. What clearly matters is political engagement itself: interested respondents talk more everywhere, but especially in private settings. What does not yet travel well from theory to pilot is the broader social-risk package. Minority positioning, cross-cutting friendships, and institutional trust all remain weak in the main models. The public-service mechanism looks more plausible, but mainly as a private-talk story. The preregistered H7 extension is also worth keeping: women talk less at work and show larger gaps on average, and the woman-by-migrant interaction points toward a smaller gap, although that estimate is still imprecise.

7 Functional-Form Robustness

The preregistration asks for OLS versions of the two ordered models as a robustness check. This is simply a way to see whether the conclusions depend heavily on the choice of ordered logit. If the same broad patterns show up when we treat the 1 to 7 scales as approximately continuous, that makes the main story look more stable. The broad picture does not change: political interest remains the clearest and most stable predictor, while the other mechanisms stay much weaker.

7.1 Private Talk OLS

term estimate std.error statistic p.value conf.low conf.high term_label
political_interest_z 0.957 0.078 12.281 0.000 0.805 1.110 Political interest
service_dissatisfaction_z 0.329 0.113 2.897 0.004 0.106 0.551 Public-service dissatisfaction
political_opinion_minority 0.317 0.386 0.821 0.412 -0.440 1.075 Political-opinion minority
institutional_trust_index_z 0.173 0.115 1.505 0.133 -0.052 0.397 Institutional trust
cross_cutting_friendships_z -0.061 0.082 -0.751 0.453 -0.222 0.099 Cross-cutting friendships

7.2 Workplace Talk OLS

term estimate std.error statistic p.value conf.low conf.high term_label
political_interest_z 0.693 0.121 5.710 0.000 0.455 0.931 Political interest
service_dissatisfaction_z 0.226 0.150 1.512 0.132 -0.067 0.520 Public-service dissatisfaction
institutional_trust_index_z 0.045 0.162 0.278 0.781 -0.273 0.363 Institutional trust
cross_cutting_friendships_z -0.114 0.117 -0.974 0.331 -0.345 0.116 Cross-cutting friendships
political_opinion_minority -0.516 0.530 -0.973 0.331 -1.554 0.523 Political-opinion minority

7.3 Talk Gap OLS

term estimate std.error statistic p.value conf.low conf.high term_label
political_opinion_minority 0.738 0.581 1.270 0.205 -0.401 1.876 Political-opinion minority
political_interest_z 0.297 0.116 2.549 0.011 0.069 0.525 Political interest
service_dissatisfaction_z 0.127 0.148 0.855 0.393 -0.164 0.418 Public-service dissatisfaction
institutional_trust_index_z 0.091 0.157 0.581 0.562 -0.217 0.400 Institutional trust
cross_cutting_friendships_z -0.010 0.108 -0.090 0.929 -0.221 0.202 Cross-cutting friendships

7.4 Classic OLS Regression Tables

These tables present the same OLS robustness models in a more standard regression layout.

Characteristic

Private talk OLS

Workplace talk OLS

Talk gap OLS

Beta

1

SE

2

Beta

1

SE

2

Beta

1

SE

2
Political interest 0.957*** 0.085 0.693*** 0.112 0.297** 0.106
Political-opinion minority 0.317 0.360 -0.516 0.500 0.738 0.474
Cross-cutting friendships -0.061 0.088 -0.114 0.116 -0.010 0.110
Institutional trust 0.173 0.119 0.045 0.158 0.091 0.150
Public-service dissatisfaction 0.329** 0.110 0.226 0.144 0.127 0.136
1

p<0.05; p<0.01; p<0.001
2

SE = Standard Error

8 Moderation Tests

The preregistered moderation checks ask whether minority positioning is conditioned by political interest and cross-cutting friendships in the workplace and gap models.

model term_label estimate conf.low conf.high p.value
Talk gap: minority × cross-cutting Minority × cross-cutting friendships 1.114 0.476 1.752 0.001
Talk gap: minority × interest Minority × political interest -0.099 -1.253 1.055 0.867
Workplace talk: minority × cross-cutting Minority × cross-cutting friendships -1.101 -2.147 -0.082 0.034
Workplace talk: minority × interest Minority × political interest -0.540 -1.738 0.672 0.368

The minority-by-interest interaction is essentially absent in the pilot. The minority-by-cross-cutting-friendships interaction is more interesting, but it runs against the intended story: among respondents who see their political views as a minority position, more cross-cutting friendships are associated with less workplace talk and a larger gap rather than less self-suppression.

9 H7 Model

H7 is estimated separately because it is asking a different question from the main mechanism models. The main models focus on political interest, minority positioning, cross-cutting friendships, trust, and service experiences. H7 instead asks whether the private-workplace gap differs for a specific social-position combination: women with a migrant background. To test that properly, the model needs gender, migrant background, and their interaction term. That is why it is clearer to estimate H7 on its own rather than fold it into the main mechanism table.

model term_label estimate conf.low conf.high p.value
Talk gap Woman 0.629 0.192 1.066 0.005
Talk gap Migrant background 0.018 -0.786 0.822 0.965
Talk gap Woman × migrant background -0.942 -2.027 0.142 0.090
Workplace talk Woman × migrant background 0.514 -1.077 2.109 0.525
Workplace talk Migrant background -0.251 -1.442 0.929 0.676
Workplace talk Woman -0.788 -1.279 -0.302 0.002

In the pilot, the clearest H7 result is the main gender difference rather than the full interaction pattern. Women report less political discussion at work and a larger private-workplace gap on average, and both estimates are reasonably precise. The woman-by-migrant-background interaction is much less certain in the workplace-talk model, but in the gap model it points in the preregistered direction: the gap is smaller for women with a migrant background than for women without that background. If we treat p < 0.10 as suggestive evidence, which is reasonable in a pilot setting, that interaction is worth keeping in view as a possible lead for the full study rather than as a settled result.

10 Optional Gap Typology

As an additional robustness check, the gap can be turned into four easy-to-read patterns instead of one numeric difference score. I use a simple cutoff of talking at least weekly (5 or above on the original 1 to 7 scale). That creates four groups: low talk in both settings, high private but low workplace talk, low private but high workplace talk, and high talk in both settings. This is useful because some readers find these patterns easier to interpret than a gap score like 2 or -1. At the same time, this is only a descriptive extension. It is not a stronger test than the main models, because it throws away some detail from the original scale in order to create simpler categories.

gap_pattern n pct
Low private / Low workplace 138 47.3
High private / High workplace 94 32.2
High private / Low workplace 46 15.8
Low private / High workplace 14 4.8
note
The multinomial specification was estimated as a sensitivity check, but the resulting category-specific coefficient tables are not especially informative in this pilot. Some categories are small, one category shows signs of quasi-separation, and the substantive takeaway does not improve on the simpler main models. For that reason, the report keeps the pattern counts above and treats the multinomial model as a background check rather than a table worth presenting in full.

11 Notes

  • poldiscol == 99 is treated as “no colleagues” and excluded from workplace and gap analyses.
  • The talk gap is defined only when respondents have colleagues and valid answers to both discussion items.
  • H2 uses political-opinion minority status as the main prereg proxy because it matches the theory of expression risk more directly than a broad minority-status indicator.
  • The results are descriptive pilot associations, not causal estimates.

12 Appendix

12.1 Sample Restrictions and Missingness

variable n_total n_missing pct_missing
workplace_political_discussion 355 63 17.7
political_discussion_gap 355 63 17.7
education_level 355 52 14.6
ideology_right 355 7 2.0
employment_status 355 2 0.6
gender_binary 355 1 0.3
migrant_background 355 1 0.3
private_political_discussion 355 0 0.0
political_interest 355 0 0.0
political_opinion_minority 355 0 0.0
cross_cutting_friendships 355 0 0.0
institutional_trust_index 355 0 0.0
service_dissatisfaction_z 355 0 0.0
age 355 0 0.0
model n pct_of_total
Private talk model 295 83.1
Workplace talk model 254 71.5
Talk-gap model 254 71.5