A Standardized Field-Survey Protocol for Visitor Conservation Willingness and Self-Reported Responsible Travel Behavior at Longmen Grottoes

Summary

This protocol standardizes a minimum-disruption visitor survey at Longmen Grottoes by linking public-area interception, anonymous questionnaire administration, predefined screening, measurement validation, model checking, and reproducibility archiving.

Abstract

Heritage-site visitor surveys must generate usable behavioral evidence without disturbing site operations or weakening conservation priorities. However, recruitment boundaries, exclusion rules, measurement checks, model diagnostics, and data archiving are often incompletely reported in heritage tourism studies. This protocol presents a minimum-disruption field-survey workflow for measuring visitor conservation willingness and self-reported responsible travel behavior at Longmen Grottoes, Luoyang, China. The workflow integrates public-area visitor interception, anonymous electronic questionnaire administration, eligibility screening, attention checks, daily raw-data locking, predefined exclusion rules, construct scoring, reliability testing, confirmatory factor analysis, regression and path-model diagnostics, sensitivity checks, and reproducibility archiving. The questionnaire separates heritage knowledge, perceived heritage value, place attachment, pro-environmental message quality, perceived crowding, facility and service support, conservation attitude, conservation willingness, and self-reported responsible travel behavior. Representative implementation showed that the workflow produced an analyzable dataset with auditable exclusions, acceptable measurement diagnostics, and interpretable model outputs. The protocol is intended for heritage-site studies that need transparent visitor-behavior evidence rather than a single satisfaction measure. It can be adapted to other heritage settings by retaining the core procedural sequence while revising site-specific constructs, survey points, language versions, behavioral items, and fieldwork restrictions according to local conservation risks and visitor composition.

Introduction

Cultural heritage sites require visitor access for public education, cultural transmission, and tourism value, but visitor use can also place pressure on fragile material, spatial, and symbolic resources¹. This tension is especially visible at high-visitor-volume sites where movement routes are limited, interpretation areas are dense, and conservation rules must be understood during a short visit. Longmen Grottoes in Luoyang, China, provides an appropriate setting for a conservation-oriented visitor-survey workflow because it combines rock-carving heritage protection, public interpretation, visitor-flow management, and conservation education within one active field environment. In this setting, visitor measurement cannot stop at satisfaction. A useful survey must document whether visitors understand heritage value, recognize protection rules, accept conservation restrictions, and report responsible conduct during or immediately after the visit.

Heritage tourism studies commonly measure perceived value, destination image, place attachment, satisfaction, or general pro-environmental intention². The operational weakness is not the use of these constructs, but the limited reporting of how field evidence is produced. Recruitment points, sampling intervals, suspension rules, attention checks, exclusion thresholds, raw-data locking, construct scoring, measurement validation, model diagnostics, and reproducibility archiving are often reported briefly or after results are known. This weakens comparison across studies and limits the usefulness of survey evidence for heritage management. The present protocol addresses this gap by fixing the workflow sequence before substantive interpretation: questionnaire preparation, pilot testing, public-area recruitment, anonymous administration, data screening, measurement checking, model interpretation, sensitivity analysis, and archiving.

The construct structure follows the practical order of a conservation-oriented heritage visit. Heritage knowledge records whether visitors understand the historical and cultural meaning of the site. Perceived heritage value captures whether visitors regard the site as worthy of protection³. Place attachment measures the personal connection formed through the visit⁴. Pro-environmental message quality records whether conservation messages are visible, understandable, and useful for guiding conduct. Perceived crowding captures density and movement pressure that may affect attention to rules⁵. Facility and service support records whether signs, barriers, routes, staff reminders, and visitor services make responsible conduct feasible. These six site-related constructs are separated from conservation attitude, conservation willingness, and self-reported responsible travel behavior so that perception, evaluation, intention, and reported conduct are not collapsed into one general positive-response score.

Conservation attitude and conservation willingness are also treated as distinct constructs. Conservation attitude refers to the visitor’s evaluative belief that Longmen Grottoes should be protected and that conservation should be prioritized. Conservation willingness refers to the visitor’s readiness to accept or support conservation-related requirements, including rule-following, inconvenience, route restrictions, and protective management⁶. Self-reported responsible travel behavior refers to conduct reported during the visit, such as following signs, avoiding harmful contact, staying within permitted areas, and reducing disturbance to the heritage environment⁷. A visitor may agree that conservation is important but still be unwilling to accept restrictions. Responsible conduct may also occur because the site environment makes compliance easy. Separating these constructs helps identify whether a management issue is closer to knowledge, message clarity, willingness, facility support, or reported conduct.

Longmen Grottoes is used as a testing context because visitor measurement must occur close to the visit while preserving low disruption, anonymity, and operational boundaries. The workflow is restricted to public visitor-service areas and post-visit or near-exit spaces, not restricted conservation zones or direct observation of protected surfaces. This boundary matters because heritage-site fieldwork takes place in active management settings where survey activity must not interrupt movement, staff duties, visitor safety, or conservation practice⁸. Fixed survey windows, systematic interception, suspension rules, predefined exclusions, and daily raw-file locking reduce undocumented field decisions and post hoc data cleaning.

The protocol is transferable at the workflow level, but the questionnaire is not intended to be copied unchanged. The components that should remain unchanged are survey-boundary definition, target-population definition, finalized questionnaire version control, systematic visitor interception, eligibility screening, attention checks, raw-data locking before exclusion, predefined data-quality rules, construct scoring, measurement validation, model diagnostics, sensitivity checks, and reproducibility archiving. The adjustable components are conservation-risk wording, behavioral items, survey points, language versions, visitor-composition variables, and management indicators. Archaeological parks may require stronger wording on surface contact and route restriction⁹. Historic towns may require items on resident disturbance and shared street-space behavior¹⁰. Museums and sacred landscapes may require gallery-flow, exhibit-protection, ritual-space, or culturally sensitive conduct items¹¹. This structure allows site-specific adaptation without weakening procedural comparability.

Representative results are used only to demonstrate whether the workflow produces usable visitor-behavior data. They are not used to claim direct observation of behavior, carrying-capacity estimation, intervention effects, or equivalence between self-reported and observed conduct. The contribution of the protocol is the standardized, auditable sequence for collecting, screening, validating, analyzing, and archiving conservation-oriented visitor-survey data under minimum-disruption heritage-site conditions.

This standardized workflow measures visitor conservation willingness and self-reported responsible travel behavior at Longmen Grottoes, Luoyang, Henan Province, China. It includes questionnaire design, pilot testing, public-area visitor interception, anonymous questionnaire administration, raw-data locking, quality screening, construct scoring, measurement checking, model diagnostics, and study-file archiving.

Protocol

The study was conducted in accordance with the Declaration of Helsinki and was approved by the SEGi Research Ethics Committee (Approval No. SEGiEC/SR/FOELPM/163/2025-2027) on 27 June 2025.Written or electronic informed consent was obtained before the questionnaire began.

NOTE: No names, identity card numbers, telephone numbers, facial images, home addresses, device identifiers, GPS coordinates, account identifiers, photographs, audio recordings, video recordings, or other directly identifiable information were collected. IP address collection was disabled before formal questionnaire release. No formal written site-management permission was obtained from the Longmen Grottoes management authority. Therefore, fieldwork was restricted to publicly accessible visitor-service areas and post-visit or near-exit spaces. Survey assistants did not enter grotto interiors, restricted conservation zones, staff-only areas, narrow passageways, crowded viewing platforms, ticket-checking queues, emergency routes, religious or commemorative spaces, or any location where recruitment could interfere with visitor movement, site operations, visitor safety, staff duties, or conservation practice. Recruitment was suspended immediately during heavy rain, emergency announcements, official crowd-control measures, staff requests, abnormal crowding, or any situation that could interfere with visitor flow, safety, or site management.

1. Define the field site, constructs, sample target, and survey window

1. Use Longmen Grottoes scenic area in Luoyang, Henan Province, China, as the field site. Define the target population as adult visitors who have completed, or are about to complete, their same-day visit to Longmen Grottoes.
2. Use three public survey points: the visitor exit area, the public rest area outside the main visiting route, and the visitor service area near the transportation connection point. Use only spaces where visitors can stop voluntarily without blocking visitor movement or interfering with site operations.
3. Suspend recruitment during heavy rain, emergency announcements, official crowd-control measures, staff requests, abnormal crowding, guided-tour congestion, blocked pathways, or any situation that could interfere with visitor flow, visitor safety, staff duties, or site management.
4. Use heritage knowledge (HK), perceived heritage value (PHV), place attachment (PA), pro-environmental message quality (PMQ), perceived crowding (PC), and facility and service support (FSW) as site-related predictors.
5. Use conservation attitude (CA) and conservation willingness (CW) as intention-related constructs. Define CA as the visitor’s belief that Longmen Grottoes should be protected and that conservation should be prioritized. Define CW as the visitor’s readiness to accept or support conservation requirements, including rule-following, inconvenience, route restrictions, and protective management.
6. Use self-reported responsible travel behavior (RTB) as the behavioral outcome. Define RTB as conduct reported during the visit, including following signs and barriers, avoiding harmful contact with heritage surfaces, staying within permitted areas, reducing disturbance to other visitors and the heritage environment, and keeping the site clean.
7. Set the planned raw sample size at 480 completed questionnaires and the minimum valid sample size at 400 questionnaires after quality screening.
8. Conduct formal data collection from 1 April 2026 to 20 May 2026 . Use two survey periods per fieldwork day: 10:00–12:00 and 14:30–17:30.
9. Figure 1 summarizes the full procedure from questionnaire design and pilot testing to public-area field collection, screening, scoring, measurement validation, model checking, sensitivity analysis, and study-file archiving.
   NOTE: Longmen Grottoes was selected because this World Heritage cultural site combines high visitor volume, rock-carving heritage conservation, public interpretation, and visitor-management demands within one active field setting¹².

2. Construct, screen, translate, and pilot the questionnaire

1. Organize the questionnaire into five sections: informed consent, eligibility screening, heritage-site perception items, conservation attitude, conservation willingness and responsible travel behavior items, and demographic and travel-related variables.
2. Use a 5-point Likert scale for all substantive perception, attitude, willingness, and behavior items: 1 = strongly disagree, 2 = disagree, 3 = neither agree nor disagree, 4 = agree, and 5 = strongly agree.
3. Include 31 substantive Likert items across nine constructs: HK, PHV, PA, PMQ, PC, FSW, CA, CW, and RTB. Use at least three observed items for each construct, with four items for PHV, CA, CW, and RTB.
4. Keep CA, CW, and RTB wording distinct. CA items assess the perceived importance, priority, and responsibility of conservation. CW items assess willingness to accept or support conservation-related requirements. RTB items assess conduct reported during the same-day visit.
5. Word each item around the Longmen Grottoes visit rather than general environmental concern. Avoid double-barreled wording, leading wording, professional conservation terminology, and wording that implies a socially desirable answer.
6. Align willingness and behavior items with behavioral-intention and perceived-control logic in applied visitor research¹³. Adapt place attachment and responsible-behavior items to the Longmen Grottoes visit¹⁴. Use heritage-site perception items to measure value perception, visitor-site relationship quality, and conservation-related interpretation specific to Longmen Grottoes¹⁵.
7. Keep the average reading length of each item below 28 characters where possible and keep the full questionnaire completion time within 6–10 min during pilot testing. Do not include open-ended free-text questions in the formal questionnaire.
8. Add screening items for age, same-day visit status, occupational role, and repeat participation. Terminate the questionnaire if the participant is younger than 18 years, did not visit Longmen Grottoes on the survey day, is a scenic-area staff member, tour guide working during the visit, travel-agency employee working during the visit, vendor working inside or near the survey area, member of the research team, or has already completed the questionnaire on the same day.
9. Add two attention-check items. Place the first after the first one-third of substantive items: “To confirm that you are reading the questions carefully, please select ‘Agree’ for this item.” Code it as passed only when the participant selects 4. Place the second after the second one-third of substantive items: “For quality-control purposes, please select ‘Disagree’ for this item.” Code it as passed only when the participant selects 2.
10. Prepare the working questionnaire in preferred language. Ask two bilingual researchers with tourism, heritage, or behavioral survey experience to compare the items with the intended English construct meanings. Resolve wording disagreements before pilot testing.
11. Conduct the pilot test from 20 March 2026 to 25 March 2026 with 30 adult visitors or recent visitors to a cultural heritage attraction. Record completion time, skipped items, misunderstood wording, repeated participant questions, item hesitation, platform-display problems, and confusion about CA, CW, or RTB wording.
12. Revise any item if more than 20% of pilot participants report that it is difficult to understand. Retain an item if its corrected item-total correlation is at least 0.30. Retain a construct if Cronbach’s alpha is at least 0.70.
13. Retain a construct with Cronbach’s alpha between 0.65 and 0.69 only if all corrected item-total correlations are at least 0.30 and no single item deletion increases Cronbach’s alpha by more than 0.03. Mark any construct in this range as marginally reliable.
14. Exclude pilot-test responses from the formal analytical dataset. Finalize the questionnaire after the revised items are checked by the two bilingual reviewers.
15. Table 1 summarizes the construct definitions, item codes, item wording, scoring direction, analytical position, and source or adaptation basis for the nine questionnaire constructs.
16. Supplementary File 1 provides the final questionnaire, informed-consent page, screening questions, attention-check items, questionnaire-platform display record, and pilot-review record.
    PAUSE POINT: Stop after the pilot test if any construct has Cronbach’s alpha below 0.65. Revise the item wording and repeat the pilot test before formal fieldwork.

3. Train survey assistants and conduct systematic intercept sampling

1. Recruit survey assistants who are not involved in scenic-area management, ticketing, tour operation, souvenir sales, visitor-flow control, conservation enforcement, or any work role that could influence visitors’ willingness to participate.
2. Assign each survey assistant a fieldwork code before training. Use personnel codes such as SA01, SA02, SA03, and SA04 in the fieldwork log instead of personal names.
3. Train all survey assistants for 2 h before formal data collection. Cover eligibility criteria, consent procedure, invitation script, refusal handling, sampling interval, travel-group selection rule, recruitment suspension rule, privacy requirements, quality-control rules, fieldwork-log completion, and the distinction among CA, CW, and RTB.
4. Ensure each survey assistant completes five supervised practice invitations before formal data collection.
5. Use the following invitation script: “Hello, we are conducting a de-identified academic survey about visitor experience, heritage protection, and responsible travel behavior at Longmen Grottoes. The questionnaire takes about 6–10 min. Participation is voluntary, and you may stop at any time.”
6. Do not mention expected hypotheses, desired response directions, model relationships, conservation-behavior scores, or any assumed positive effect during recruitment.
7. At the beginning of each survey window, randomly select a starting number between 1 and 5. Approach the visitor corresponding to the selected starting number, and then approach every fifth adult visitor passing the selected survey point after completing, or nearly completing, the visit.
8. Reset the counting sequence after each completed questionnaire, refusal, or ineligible visitor.
9. Invite only one person from the same travel group. When multiple eligible visitors from the same travel group volunteer, invite the person whose birthday is closest to the survey date.
10. End the invitation immediately if the visitor declines, appears rushed, is caring for children or older companions, is blocking visitor flow, is listening to a guided explanation, is entering a crowded route segment, or is in any situation where recruitment may cause inconvenience.
11. Maintain a fieldwork log for each survey window. Record survey date, time window, survey point, survey assistant code, visitors approached, ineligible visitors, refusals, started questionnaires, completed questionnaires, suspended periods, reason for suspension, and sampling deviations.
12. Use the fieldwork log to calculate approach rate, refusal rate, completion rate, and valid-response rate.
13. Continue fieldwork until 480 raw questionnaires are collected. Ensure that each survey point contributes at least 120 raw questionnaires.
14. Monitor gender, age group, first-time visit status, and travel mode at the end of each survey day. Do not reject eligible visitors to force demographic balance. Keep first-time visitors between 45% and 70% and organized group-tour participants below 45% of the valid sample where possible. Treat these values as monitoring ranges, not quotas.
15. Table 2 presents the fieldwork schedule, public survey points, sampling interval, sample targets, suspension rules, and fieldwork-recording variables.
16. Supplementary File 2 provides the fieldwork-log form, role-assignment record, data-management log, exclusion-log form, and codebook structure.
    NOTE: Record any deviation from the sampling interval, survey point, time window, or suspension rule in the fieldwork log on the same day.

4. Administer the questionnaire and check daily records

1. Show the informed-consent page before the questionnaire begins. Ask each participant to confirm that they are at least 18 years old, have read the consent statement, and agree to participate voluntarily.
2. Administer the questionnaire through an electronic questionnaire platform on a tablet or on the participant’s mobile phone. Present one item or one short item group per screen and require a response for each substantive Likert item.
3. Explain only the literal meaning of a word or phrase if clarification is requested. Do not suggest, imply, or confirm any response direction.
4. Record start time, end time, completion time, date, survey point, time window, and survey assistant code. Do not collect free-text comments during the formal questionnaire.
   CHECKPOINT: At the end of each survey day, confirm that all completed questionnaires contain start time, end time, survey point, time window, and survey assistant code.

5. Export, screen, and score the dataset

1. Export the raw questionnaire file in .xlsx format at 20:00 on each survey day. Name each file using the format LongmenSurvey_raw_YYYYMMDD.xlsx and store it in a password-protected folder accessible only to authorized study personnel.
2. Maintain a role-assignment log using personnel codes rather than names. Record the personnel code responsible for each survey session, daily export, raw-file storage, raw-file locking, exclusion coding, independent screening verification, statistical analysis, and final dataset freeze.
3. Merge all daily raw files after formal data collection ends. Assign each case a unique respondent ID using the format LGV0001 to LGV0480. Save one unedited master file as LongmenSurvey_raw_locked.xlsx, and record the lock date, file size, and SHA-256 checksum in the data-management log.
4. Conduct all screening, scoring, and analysis in copied working files. Do not edit the locked raw master file.
5. Exclude cases with eligibility failure, including age below 18 years, no same-day visit, excluded occupational role, or repeated same-day participation. Exclude cases that fail either attention-check item.
6. Exclude cases completed in less than 180 s, cases with the same Likert response across 15 or more consecutive substantive items, cases with a standard deviation below 0.25 across all substantive Likert items, and confirmed duplicate cases.
7. Review cases completed in more than 1,800 s. Retain a long-duration case only if it has no attention-check failure, straight-lining, eligibility contradiction, duplicate response pattern, or other fieldwork problem.
8. Check duplicate cases using substantive Likert items, demographic variables, survey point, time window, and completion time rounded to the nearest second. If a complete duplicate is found, retain the earlier case and exclude the later case.
9. Create an exclusion variable with the categories: eligible, eligibility failure, attention-check failure, fast completion, long-duration review failure, straight-lining, duplicate response, and other fieldwork problem. Retain only cases coded as eligible. Continue fieldwork under the same rules if fewer than 400 valid cases remain after screening.
10. Complete independent screening verification before construct scoring. The first reviewer applies all exclusion rules, and the second reviewer checks eligibility status, attention-check results, completion time, straight-lining, duplicate decisions, and final exclusion codes against the locked raw file.
11. Code all Likert items from 1 to 5. Reverse-code negatively worded items using reversed score = 6−original score.
12. Code demographic and travel variables, including gender, age group, education, monthly personal income, first-time visit, travel mode, survey point, and time window. Record all coding rules in the codebook.
13. Calculate each construct score as the arithmetic mean of its retained items and keep all construct scores on the original 1–5 scale. Do not impute item-level missing values because all substantive items are required fields.
14. Table 3 summarizes the eligibility, attention-check, and quality-screening rules applied before construct scoring.
15. Table 4 presents the demographic coding, travel-variable coding, reverse-coding rule, exclusion-status coding, and construct-score calculation plan.
    CHECKPOINT: Proceed to measurement testing only after the final analytical dataset contains at least 400 valid questionnaires, all retained cases have complete construct scores, and independent screening verification has been completed.

6. Assess reliability, measurement structure, and common-method risk

1. Calculate Cronbach’s alpha and corrected item-total correlation for each multi-item construct. Treat Cronbach’s alpha ≥0.70 as acceptable. Treat Cronbach’s alpha between 0.65 and 0.69 as marginally reliable only when all corrected item-total correlations are ≥0.30 and no single item deletion increases alpha by more than 0.03.
2. Retain items with corrected item-total correlation ≥0.30. Remove an item only if its corrected item-total correlation is below 0.30 and its removal increases Cronbach’s alpha by at least 0.03. Mark perceived crowding as marginally reliable if its alpha remains between 0.65 and 0.69 after item review.
3. Conduct confirmatory factor analysis for the nine-factor measurement model using all retained substantive items. Report CFI, TLI, RMSEA, SRMR, standardized factor-loading range, composite reliability, and average variance extracted.
4. Treat the measurement model as acceptable when CFI and TLI are ≥0.90 and RMSEA and SRMR are ≤0.08. Retain items with standardized factor loading ≥0.50. Review any item with loading below 0.50 or large residual correlation with a conceptually similar item.
5. Assess discriminant validity using latent construct correlations and HTMT. Flag construct pairs if HTMT exceeds 0.85, or 0.90 for conceptually adjacent constructs. Review CA, CW, and RTB carefully if they show excessive latent overlap.
6. Compare the nine-factor model with a one-factor model as a common-method risk screen. Conduct Harman’s single-factor test only as a descriptive screen, and do not use it as proof that common-method bias¹⁶ is absent.
7. Conduct exploratory factor analysis only as a secondary item-clustering diagnostic, using principal axis factoring with oblimin rotation.
8. Table 5 summarizes the pre-specified thresholds used for reliability, item retention, CFA fit, discriminant validity, common-method screening, correlation review, model diagnostics, indirect-effect testing, and sensitivity checks.
9. Report the full item-level reliability results, standardized CFA loadings, composite reliability, AVE values, HTMT matrix, one-factor comparison model, and exploratory item-clustering diagnostics in Supplementary File 3.
   PAUSE POINT: Stop after measurement testing if more than one construct has Cronbach’s alpha below 0.65, if CA, CW, and RTB cannot be separated in the measurement model, or if any HTMT value exceeds the pre-specified threshold.

7. Generate descriptive diagnostics and estimate models

1. Calculate frequency and percentage for categorical variables. Calculate mean, standard deviation, median, minimum, maximum, skewness, and kurtosis for all construct scores.
2. Treat absolute skewness below 2.00 and absolute kurtosis below 7.00 as acceptable for the planned models. Flag a construct if more than 60% of valid participants select the same rounded response category or if its standard deviation is below 0.50.
3. Calculate Pearson correlations among construct scores. Treat correlations above 0.70 as possible construct redundancy and recheck item coding, reverse coding, and construct calculation before modeling.
4. Standardize continuous construct scores and dummy-code categorical controls. Use age group, gender, education, monthly personal income, first-time visit, and travel mode as control variables.
5. Estimate three regression models. Regress CA on HK, PHV, PA, PMQ, PC, FSW, and controls. Regress CW on HK, PHV, PA, PMQ, PC, FSW, CA, and controls. Regress RTB on HK, PHV, PA, PMQ, PC, FSW, CA, CW, and controls.
6. Record standardized coefficient, standard error, t value, p value, 95% confidence interval, R², adjusted R², and VIF for each regression model.
7. Estimate a path model following the construct sequence: site-related predictors → CA → CW → RTB. Test indirect effects from site-related predictors to CW and RTB through CA, and from CA to RTB through CW, using 5,000 bootstrap samples.
8. Report standardized direct effects, indirect effects, total effects, bootstrap 95% confidence intervals, and model fit indices. Treat p <0.05 as statistically significant and p values between 0.05 and 0.10 as marginal.
9. Do not describe non-significant paths as confirmed effects, and do not infer mediation without indirect-effect testing.
10. Present only the summary measurement and model diagnostics in the main text. Report the complete correlation matrix, regression coefficients, path-model estimates, indirect effects, and total effects in Supplementary File 3.

8. Assess model diagnostics and sensitivity

1. Calculate the variance inflation factor for each predictor in each regression model. Treat VIF < 3.00 as acceptable, inspect values between 3.00 and 5.00, and justify any model retained with VIF > 5.00.
2. Flag cases with absolute standardized residuals greater than 3.00. Calculate Cook’s distance for each regression model and flag cases with Cook’s distance greater than 4/n, where n is the number of valid cases included in the model.
3. Treat R² between 0.15 and 0.35 as plausible for cross-sectional visitor behavior data. Recheck the data if all hypothesized paths are significant, all standardized coefficients exceed 0.40, or any R² exceeds 0.60.
4. Repeat the three regression models after removing cases with completion time longer than 1,800 s, excluding organized group-tour participants, recalculating construct scores using median item scores, and excluding cases flagged by Cook’s distance.
5. Repeat the path model after excluding cases flagged by Cook’s distance.
6. Treat results as stable if the main coefficient directions remain unchanged and the key significant or marginal paths remain in the same direction. Report any coefficient-direction change, loss of significance, or large adjusted R² change in the representative results.
7. Report the complete sensitivity-check outputs in Supplementary File 3.

9. Run the analysis and archive the study files

1. Conduct data cleaning in copied working files and complete statistical analysis using the R version and package versions listed in the Table of Materials.
2. Run the analysis script using only the de-identified valid dataset. Set the random seed to 20260401 before resampling, bootstrap estimation, or sensitivity checks.
3. Save the R session information using sessionInfo() after the analysis script has finished.
4. Export the de-identified valid dataset, codebook, exclusion log, fieldwork-log summary, role-assignment record, analysis script, session information, complete model outputs, and figure and table source data using the file names listed in the codebook.
5. Verify that the exported files reproduce the final valid sample size, construct scores, reliability results, CFA indices, HTMT values, regression estimates, path-model estimates, indirect effects, and sensitivity-check results.
6. Create one study archive folder after all analyses have been completed. Place the de-identified valid dataset, codebook, exclusion log, fieldwork-log summary, role-assignment record, analysis script, sessionInfo() output, complete model outputs, and figure and table source data in the folder.
7. Store the locked raw dataset separately in encrypted institutional storage for 5 years after publication. Do not place locked raw export files, platform-generated identifiers, IP addresses, GPS coordinates, device identifiers, staff-identifiable field notes, or potentially identifying free-text information in the study archive folder.
8. Confirm that every variable in the de-identified dataset is defined in the codebook and that every excluded case is linked to one exclusion category in the exclusion log. Record the archive date, file version, file size, and checksum for the de-identified valid dataset in the data-management log.
9. Deposit the de-identified reproducibility package in a permanent public repository before final publication. The package should include the de-identified item-level dataset, final questionnaire, informed-consent page, screening questions, attention-check items, codebook, fieldwork-log summary, exclusion log, data-management log, role-assignment record, analysis script, session information, complete model outputs, and figure and table source data. Deposit the package in Zenodo (https://doi.org/10.5281/zenodo.21640066).
10. Supplementary File 3 and Supplementary Tables 1–7 provides the analysis script, sessionInfo() output, measurement-output workbook, model-output workbook, sensitivity-check outputs, and figure/table source-data checklist.
    PAUSE POINT: Do not share any analytical file until direct identifiers, platform-generated identifiers, location traces, staff-identifiable notes, and potentially identifying text have been removed.

Results

Participant retention and fieldwork profile

A total of 480 questionnaires were obtained during formal fieldwork at Longmen Grottoes. After eligibility screening and data-quality checks, 430 questionnaires were retained for analysis, giving a valid-response rate of 89.6%. The final valid sample exceeded the pre-specified minimum requirement of 400 valid questionnaires. Excluded cases included eligibility failures, failed attention checks, fast completion, straight-lining or near-invariant response patterns, duplicate responses, and other fieldwork problems. When one questionnaire met more than one exclusion condition, the earliest applicable protocol rule was used as the primary exclusion category. The screening flow from 480 raw questionnaires to 430 valid cases is shown in Figure 2.

The retained cases were distributed across the three planned public survey points. The visitor exit area contributed 139 valid cases, the public rest area outside the main visiting route contributed 146 cases, and the visitor service area near the transportation connection point contributed 145 cases. First-time visitors accounted for 55.6% of valid cases, and repeat visitors accounted for 44.4%. Organized group-tour participants accounted for 23.0%, which remained below the pre-specified monitoring boundary. The fieldwork distribution and visitor profile are shown in Table 6.

Construct distribution and measurement diagnostics

The construct scores showed sufficient dispersion for analysis. Mean scores ranged from 3.258 to 3.415, and standard deviations ranged from 0.837 to 0.892. No construct showed a strong ceiling effect based on the pre-specified dispersion checks. Perceived crowding was retained in its original direction, so a higher score indicated stronger perceived crowding rather than a more positive visitor experience.

Cronbach’s alpha ranged from 0.699 for perceived crowding to 0.812 for self-reported responsible travel behavior. All constructs met either the acceptable reliability threshold or the marginal-reliability rule. Perceived crowding was reported as marginally reliable because its alpha was slightly below 0.70. It was retained because the corrected item-total correlations met the protocol threshold and the construct was needed to represent visitor-pressure conditions at a high-traffic heritage site.

The nine-factor confirmatory factor analysis supported the planned measurement structure. The model fit indices were acceptable: χ² (398) = 712.4, CFI = 0.932, TLI = 0.918, RMSEA = 0.043, and SRMR = 0.052. Standardized factor loadings ranged from 0.56 to 0.84. Composite reliability values ranged from 0.702 to 0.842. Average variance extracted ranged from 0.453 to 0.575, with the lowest value occurring for perceived crowding.

Discriminant-validity checks did not indicate severe construct overlap. The maximum HTMT value was 0.642, below the pre-specified review threshold. Conservation attitude, conservation willingness, and self-reported responsible travel behavior remained distinguishable in the measurement model. The first unrotated factor explained 18.7% of the total variance, below the 40% common-method screening threshold. This result was treated only as a screening result and not as evidence that common-method bias was absent. The strongest construct correlation was between conservation attitude and conservation willingness, r = 0.432, and no construct pair exceeded the 0.70 redundancy threshold. The main measurement diagnostics are summarized in Table 7. Full item-level reliability results, CFA loadings, CR, AVE, HTMT, and inter-construct correlations are provided in Supplementary Table 1, Supplementary Table 2, and Supplementary Table 3.

Model outputs and sensitivity checks

The protocol produced interpretable regression and path-model outputs without serious multicollinearity. The three regression models showed moderate explanatory power. The conservation attitude model had R² = 0.267 and adjusted R² = 0.224. The conservation willingness model had R² = 0.333 and adjusted R² = 0.292. The self-reported responsible travel behavior model had R² = 0.275 and adjusted R² = 0.228. These values were within the pre-specified plausible range for cross-sectional visitor-behavior data.

The path model followed the planned sequence: site-related predictors → conservation attitude → conservation willingness → self-reported responsible travel behavior. Model fit was acceptable: CFI = 0.927, TLI = 0.904, RMSEA = 0.049, and SRMR = 0.041. The core path estimates were interpretable. Conservation attitude was positively associated with conservation willingness, β = 0.276. Conservation attitude and conservation willingness were positively associated with self-reported responsible travel behavior, β = 0.232 and β = 0.271, respectively. The indirect effect from conservation attitude to self-reported responsible travel behavior through conservation willingness was 0.075, with a bootstrap 95% CI of 0.036–0.122. The full regression coefficients, path estimates, indirect effects, total effects, and model-output files are provided in Supplementary File 3, Supplementary Table 4, and Supplementary Table 5.

The regression diagnostics did not indicate serious multicollinearity. VIF values ranged from 1.012 to 1.422, below the pre-specified threshold of 3.00. Influential-case diagnostics did not justify removing cases from the primary models. Cases flagged by Cook’s distance were retained in the primary analysis and evaluated in sensitivity checks.

The sensitivity checks supported the stability of the protocol outputs. No valid questionnaire exceeded the 1,800 s completion-time threshold. After organized group-tour participants were removed, after median-based construct scoring was used, and after Cook’s-distance-flagged cases were excluded, the main coefficient directions were retained. Across sensitivity specifications, the standardized coefficient for conservation attitude → conservation willingness ranged from 0.261 to 0.290, conservation attitude → self-reported responsible travel behavior ranged from 0.198 to 0.260, and conservation willingness → self-reported responsible travel behavior ranged from 0.243 to 0.298. Complete sensitivity-check outputs are provided in Supplementary Table 6.

Summary of protocol outputs

The workflow produced a valid analytical sample above the pre-specified minimum sample requirement, with completed questionnaires distributed across all planned survey points. The retained sample included variation in visitor background and travel mode, and the organized group-tour proportion remained within the monitoring boundary. The measurement checks supported the use of construct-level scores: most constructs met the reliability threshold, perceived crowding was retained as marginally reliable, the nine-factor CFA showed acceptable fit, and discriminant-validity checks supported separation among conservation attitude, conservation willingness, and self-reported responsible travel behavior.

The model outputs were used to demonstrate that the protocol could generate analyzable and auditable visitor-behavior evidence. The regression and path-model results were interpretable, the indirect-effect test was available for the planned attitude–willingness–behavior sequence, and the sensitivity checks did not change the direction of the core associations. These results support the procedural use of the workflow for minimum-disruption heritage-site visitor surveys rather than serving as direct observational evidence of visitor behavior. The de-identified dataset, questionnaires, supporting study documents, analysis script, complete model outputs, and source data for all figures and tables have been deposited in Zenodo (DOI: https://doi.org/10.5281/zenodo.21640066).

Figure 1: Protocol workflow for the Longmen Grottoes visitor survey. The workflow shows the sequence used in this protocol, from questionnaire construction, pilot testing, public-area visitor interception, questionnaire administration, and raw-data locking to quality screening, construct scoring, measurement validation, model estimation, sensitivity checks, and study-file archiving. Please click here to view a larger version of this figure.

Figure 2: Questionnaire screening flow for the final analytical sample. The diagram shows the transition from 480 raw questionnaires to 430 valid questionnaires after eligibility screening, attention-check review, completion-time screening, straight-lining detection, duplicate-response inspection, and fieldwork-related quality checks. Exclusions are counted by primary exclusion reason when more than one warning condition is present. Please click here to view a larger version of this figure.

Table 1: Construct map and questionnaire structure for the Longmen Grottoes visitor survey. This table summarizes the construct definitions, abbreviations, item codes, number of items, scoring direction, analytical position, and source or adaptation basis used to measure visitor conservation willingness and self-reported responsible travel behavior. Full item wording is provided in Supplementary File 1. All substantive items were scored from 1 = strongly disagree to 5 = strongly agree. PC was retained in its original direction; a higher PC score indicates stronger perceived crowding. Please click here to download this Table.

Table 2: Fieldwork schedule and systematic intercept sampling plan. This table presents the formal data-collection period, public survey points, daily time windows, sampling interval, raw sample target, minimum valid sample requirement, visitor-composition monitoring ranges, suspension rules, and fieldwork-recording variables. Please click here to download this Table.

Table 3: Eligibility, attention-check, and quality-screening rules. This table presents the eligibility criteria, attention-check rules, completion-time thresholds, straight-lining rules, duplicate-response checks, exclusion codes, and retained-data decisions applied before construct scoring. Please click here to download this Table.

Table 4: Variable coding and construct-score calculation plan. This table defines the coding rules for respondent ID, demographic variables, travel-related variables, survey records, attention-check variables, Likert-scale items, construct scores, and exclusion status. All construct scores remain on the original 1–5 scale. Item-level missing-value imputation was not used because all substantive questionnaire items were required fields. Please click here to download this Table.

Table 5: Pre-specified diagnostic thresholds used before model interpretation. This table reports the thresholds for reliability, item retention, confirmatory factor analysis, discriminant validity, common-method screening, distributional checks, correlation review, multicollinearity, influential-case diagnostics, model plausibility, significance testing, and indirect-effect interpretation. CFA = confirmatory factor analysis; CFI = comparative fit index;
TLI = Tucker-Lewis index; RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual; CR = composite reliability; AVE = average variance extracted; HTMT = heterotrait-monotrait ratio; VIF = variance inflation factor. Please click here to download this Table.

Table 6: Fieldwork distribution and demographic characteristics of the valid sample. This table summarizes the distribution of valid questionnaires by survey point, time window, survey period, gender, age group, education, monthly personal income, visit status, and travel mode. Percentages were calculated using the final analytical sample, n = 430. Percentages may not sum to 100.0 because of rounding. Please click here to download this Table.

Table 7: Summary of construct distribution, measurement diagnostics, model outputs, and sensitivity checks. This table summarizes construct-level descriptive statistics, Cronbach’s alpha values, CFA fit indices, composite reliability, average variance extracted, HTMT, common-method screening results, regression model fit, path-model fit, indirect-effect testing, multicollinearity diagnostics, and sensitivity-check stability. Complete item-level reliability results, CFA loadings, HTMT matrix, inter-construct correlations, regression coefficients, path estimates, indirect effects, total effects, and sensitivity-check outputs are provided in Supplementary File 3. HK = heritage knowledge; PHV = perceived heritage value; PA = place attachment; PMQ = pro-environmental message quality; PC = perceived crowding; FSW = facility and service support; CA = conservation attitude; CW = conservation willingness; RTB = self-reported responsible travel behavior. PC was retained in its original direction; higher PC indicates stronger perceived crowding. Please click here to download this Table.

Supplementary File 1: Questionnaire and consent materials. This file provides the final questionnaire, informed-consent page, eligibility-screening questions, attention-check items, questionnaire-platform display record, and pilot-review record.Please click here to download this file.

Supplementary File 2: Fieldwork and data-management records. This file provides the fieldwork-log form, role-assignment record, data-management log, exclusion-log form, and codebook structure.Please click here to download this file.

Supplementary File 3: Analysis and reproducibility materials. This file provides the de-identified analysis dataset, analysis script, sessionInfo() output, item-level reliability outputs, CFA outputs, HTMT matrix, inter-construct correlation matrix, regression outputs, path-model outputs, indirect-effect outputs, sensitivity-check outputs, complete model-output workbook, and figure/table source-data checklist.Please click here to download this file.

Supplementary Table 1: Item-level descriptive statistics and reliability outputs. All corrected item-total correlations were ≥0.30. PC = perceived crowding.Please click here to download this file.

Supplementary Table 2: Standardized CFA loadings, CR, and AVE. CFA = confirmatory factor analysis; CR = composite reliability; AVE = average variance extracted.Please click here to download this file.

Supplementary Table 3: Inter-construct correlation matrix. HK = heritage knowledge; PHV = perceived heritage value; PA = place attachment; PMQ = pro-environmental message quality; PC = perceived crowding; FSW = facility and service support; CA = conservation attitude; CW = conservation willingness; RTB = self-reported responsible travel behavior.Please click here to download this file.

Supplementary Table 4: Complete regression outputs. Controls included gender, age group, education, monthly personal income, first-time visit status, and travel mode. VIF = variance inflation factor.Please click here to download this file.

Supplementary Table 5: Path-model and indirect-effect outputs. CA = conservation attitude; CW = conservation willingness; RTB = self-reported responsible travel behavior.Please click here to download this file.

Supplementary Table 6: Sensitivity checks for the main model outputs. CA = conservation attitude; CW = conservation willingness; RTB = self-reported responsible travel behavior.Please click here to download this file.

Supplementary Table 7: Figure and table source-data checklist. Please click here to download this file.

Discussion

This protocol addresses a methodological problem in heritage tourism research: visitors may agree with heritage protection in principle, but their reported behavior during a visit is shaped by interpretation messages, route guidance, crowding pressure, facility support, and conservation rules at the same time. For Longmen Grottoes, the protocol therefore does not treat self-reported responsible travel behavior as a detached outcome. It places site-related perceptions, conservation attitude, conservation willingness, and self-reported responsible travel behavior within one standardized field-survey sequence. The main contribution is procedural rather than only substantive: the protocol links public-area interception, anonymous questionnaire administration, pre-specified screening, construct scoring, measurement diagnostics, regression and path modeling, sensitivity checks, and reproducibility archiving in a fixed order.

The measurement design stays close to the actual visit. The questionnaire asks what visitors understood about Longmen Grottoes, how they valued the site, whether they felt attached to it, how they read protection messages, whether they experienced crowding, and whether facilities and services helped them follow site rules. This site-specific structure is preferable to asking only about general environmental concern because responsible behavior at a heritage site is partly shaped by the meaning visitors attach to that site and by how that meaning is activated during the visit. Recent heritage research has similarly linked responsible visitor responses with authenticity, identity, attachment, and heritage meaning rather than treating them as detached personal traits^17,18. The protocol also separates conservation attitude, conservation willingness, and self-reported responsible travel behavior. Conservation attitude refers to believing that protection is important; conservation willingness refers to readiness to accept or support conservation requirements; self-reported responsible travel behavior refers to reported conduct during the visit. This distinction reduces the risk of treating approval of conservation as equivalent to behavioral compliance.

The representative results suggest that the measurement structure is workable without being artificially smooth. Most constructs showed acceptable internal consistency, and the confirmatory factor analysis, discriminant-validity checks, and common-method screens supported the use of separate construct scores. Perceived crowding remained close to, but slightly below, the conventional 0.70 reliability threshold. This should not be read simply as measurement failure. Crowding is a concrete but unevenly experienced condition: visitors passing through narrow viewing areas during a busy period may judge crowding differently from visitors arriving during quieter periods or moving independently. Retaining perceived crowding as a marginal construct keeps this field condition visible rather than removing it only to make the reliability table cleaner. The path-model logic also helps clarify interpretation of the representative results. Pro-environmental message quality may first help visitors understand why a rule matters and why compliance is legitimate, rather than shifting self-reported behavior directly. Work on heritage interpretation has similarly suggested that interpretation can shape pro-environmental responses through cognitive and affective mechanisms¹⁹.

The protocol also identifies what it does not measure. Longmen Grottoes visitors may receive conservation information from fixed signs, route reminders, guide explanations, staff prompts, audio guides, or digital content, but the present questionnaire measures overall pro-environmental message quality rather than separating these channels. Future applications could add a guide-mediated or digital-interpretation module while retaining the same screening, scoring, and diagnostic sequence. Research comparing cultural and natural World Heritage settings shows that guide interpretation can influence tourists’ pro-environmental behavior, although the strength of the relationship varies across site type and visitor experience²⁰. Crowding requires a similar boundary. The protocol measures perceived crowding, not objective density. This is appropriate for a visitor questionnaire because perceived crowding may affect comfort, attention, and willingness to follow rules, but it cannot replace spatial-temporal crowd monitoring. Visitor-flow and spatial-distribution studies at World Heritage sites show that crowding can be examined more directly through movement patterns and density records²¹. The survey protocol should therefore be understood as one layer of evidence, not as a carrying-capacity assessment by itself.

Another strength of the protocol is that data screening occurs before construct scoring and before model interpretation. This order reduces the risk of cleaning the dataset in response to preferred statistical results. Some site-related variables may become weak or non-significant in the final behavior model, and some field-sensitive constructs may show only marginal reliability, but such findings are retained rather than treated as errors. This is important for a protocol article because the procedure should be able to handle uneven field data, not only an idealized pattern. The use of self-reported responsible travel behavior is also practical but limited. It allows data collection across several public survey points without following visitors, filming behavior, or recording identifiable movement traces. However, self-reported pro-environmental behavior can be affected by social desirability, especially when the behavior is tied to public norms and moral expectations. Recent tourism research argues that this risk should be assessed at the level of specific behaviors rather than treated as one uniform bias across all pro-environmental measures²². For this reason, the RTB items were written as concrete visit behaviors, including following signs and barriers, avoiding harmful contact with heritage surfaces, reducing disturbance, and disposing of waste properly.

The protocol is most applicable to heritage sites that need structured visitor evidence but cannot use intrusive observation, video recording, movement tracking, or identifiable behavioral monitoring. It can be adapted to other cultural heritage settings by modifying the site-specific constructs, survey points, language versions, and fieldwork restrictions while retaining the same core sequence. Its limitations should also be clear. It was applied to one cultural heritage site and one defined data-collection window; seasonal changes, holiday crowding, route restrictions, weather conditions, ticketing arrangements, interpretation updates, and temporary management policies may affect visitor responses. The regression and path models are cross-sectional, so positive associations among conservation attitude, conservation willingness, and self-reported responsible travel behavior should not be interpreted as causal effects. When the research aim is to test whether a new interpretation message, visitor-routing strategy, or protection intervention changes actual behavior, field experiments or quasi-experimental field studies would provide stronger evidence²³. Within these boundaries, the protocol offers a reproducible way to move from on-site visitor interception to interpretable outputs, supporting a more grounded question: under what site conditions does support for conservation become conservation willingness and self-reported responsible travel behavior?

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
car (Regression diagnostic package)	Comprehensive R Archive Network	version 3.1–5	Used to calculate variance inflation factors and support regression diagnostics.
dplyr	Comprehensive R Archive Network	version 1.2.1	Data manipulation package; Used to filter valid cases, recode variables, create exclusion codes, calculate construct scores, and summarize fieldwork and model variables.
Electronic questionnaire platform	Wenjuanxing, Changsha Ranxing Information Technology Co., Ltd., Changsha, China		Web-based platform; registered-account access; accessed April–May 2026; Used to build the electronic questionnaire, set required-response fields, insert screening and attention-check items, generate the survey QR code, record completion time, and export raw responses in .xlsx format.
Field-administration tablet	Apple Inc., Cupertino, CA, USA		iPad, 10th generation; 10.9-inch display; 64 GB; Wi-Fi model; Used when participants completed the questionnaire on a research-team device at the survey point.
ggplot2	Comprehensive R Archive Network	version 4.0.3	Figure and diagnostic-plot package; Used to prepare diagnostic plots and figure source data when graphical statistical output was required.
lavaan	Comprehensive R Archive Network	version 0.7–2	Confirmatory-factor-analysis and path-model package; Used for the nine-factor confirmatory factor analysis, model-fit indices, standardized factor loadings, path model, and bootstrap indirect effects.
lm.beta	Comprehensive R Archive Network	version 1.7–3	Standardized regression package; Used to obtain standardized coefficients from linear regression models.
Microsoft Excel	Microsoft Corporation, Redmond, WA, USA	2021	Used for daily raw-file inspection, file naming, exclusion-log preparation, codebook preparation, and manual verification of .xlsx exports.
openxlsx	Comprehensive R Archive Network	version 4.2.8.1	Spreadsheet export package; Used to export cleaned datasets, codebooks, exclusion logs, reliability tables, correlation tables, regression outputs, and sensitivity-check tables.
Participant mobile-device access	Participant-owned smartphone or tablet	NA	Used when participants completed the questionnaire by scanning the QR code or opening the survey link on their own device.
psych	Comprehensive R Archive Network	version 2.6.5	Reliability and exploratory-factor-analysis package; Used to calculate Cronbach’s alpha, corrected item-total correlations, descriptive statistics, Kaiser-Meyer-Olkin value, Bartlett’s test, exploratory factor analysis, and Harman’s single-factor screen.
R Statistical computing environment	R Foundation for Statistical Computing, Vienna, Austria	version 4.6.1	Used for reliability testing, measurement diagnostics, descriptive statistics, regression analysis, path analysis, sensitivity checks, and output export.
readxl	Comprehensive R Archive Network	version 1.5.0	Spreadsheet import package; Used to import .xlsx raw data, codebooks, and exclusion logs into R.
semTools (Structural-equation-modeling utility package)	Comprehensive R Archive Network	version 0.5–9	Used to calculate composite reliability, average variance extracted, and HTMT values from the fitted measurement model.