Tourism Network Attention Variation of Chinese Cities under the COVID-19 Pandemic

Abstract

At the end of 2019, the COVID-19 pandemic broke out globally and had a tremendous impact on tourism development in countries around the world. The rapid shift of tourism from “over-tourism” to “under-tourism”, threatening the future of the global economy and society, has generated considerable interest from academia and the policy community, but the impact of COVID-19 on tourism variation remains untested by empirical evidence. Based on the daily Baidu Index of 247 prefecture-level cities in China from 2018 to 2021, this study assessed the treatment effect of COVID-19 on tourism and analyzed its dynamic characteristics using the regression-discontinuity-design (RDD) method combined with tourism network attention (TNA) data. The results show that after the outbreak of the COVID-19 pandemic, the level value of TNA dropped significantly by 2.12 (p < 0.10), and the difference value of TNA (TNA_diff) dropped significantly by 10.77 (p < 0.01), indicating that COVID-19 has a negative causal effect on tourism development, and its impact is more pronounced in major tourist source cities, with a coefficient of −14.91 (p < 0.01) corresponding to −4.57 (p < 0.01) for non-major tourist source cities when the dependent variable TNA_diff. The identification of dynamic effects further confirms that the negative impact of the pandemic on tourism network attention is fluctuating and persistent during the study period, with the two major “golden weeks” and peak season being the most severe. Compared to 2020, the TNA has generally shown an upward trend since 2021, indicating signs of a rebound in the vitality of resident tourism, which is conducive to the healthy development of the tourism market.

1. Introduction

The emergence of the new coronavirus pneumonia pandemic (COVID-19) has had a huge impact on tourism development in countries around the world [1,2]. For example, from 23 to 26 January 2020, the hotel occupancy rate in China fell by nearly 70% from a year earlier [3]. At the same time, according to the calculation of the tourism network attention index later in this study, the average value in 2019 was 40.92, and it dropped to 31.90 in 2020 due to the impact of the pandemic. The impact of the pandemic is intuitively reflected. As a fast-growing industry, fluctuations in tourism will also be rapidly transmitted to the economic, cultural, and social development sectors [4]. In mid-2019, the mass media was still debating over-tourism, with concerns that the concentration of over-tourism could lead to conflicts and complaints between tourists and residents [5]. But now, in an instant, everything had changed. Global concerns about tourism have shifted from over-tourism to under-tourism, threatening the future of the global economy and society.

However, the impact is not limited to China. The observed spread of COVID-19 around the world has led to strict domestic and international travel restrictions. Most countries around the world have restricted or completely suspended international travel. Of course, it should also be noted that China has stricter rules than countries in Europe and America, which means that the case of China can hardly be referred to other countries without restrictions. However, as a new way of monitoring tourism development, by expanding the scope of research on the tourism network attention of Chinese cities, some observable cases could also be made available to other countries.

For the tourism industry, the duration of the impact of COVID-19 is very different from other crises. The impact of COVID-19 was once considered temporary, as with other emergencies, but the global pandemic situation and COVID-19 mutations have led to the spreading and frequent recurrence of pandemics. This causes the impact of COVID-19 to linger for a long time, greatly affecting the recovery process of tourism development. Considering the damage to the real economy caused by the pandemic, households will lose sustainable income and employment opportunities. This will indirectly affect the development of tourism. How to assess the impact of the crisis as soon as possible is the focus of many scholars [2]. In previous studies, scholars have assessed the impact of SARS and other shocks on subsequent tourism development [6,7]. These analyses are usually conducted two to five years after the SARS event, but there is a lack of timely empirical analysis. In particular, a large number of literary works have a great influence on the development of the tourism industry, such as the number of tourists, tourism revenue, industry prospects, etc. [8,9,10,11,12]. In recent years, COVID-19 has attracted widespread attention from scholars and the mass media at home and abroad. However, there is still little literature on this subject. They argued, inter alia, that the crisis has a pandemic tendency and will bring a decline in the economy and tourism over a period of time through normative rather than empirical analysis [13,14,15]. Or some research notes, such as Yang et. al. [2], use the DSGE method to expand the discussion of models. The importance of tourism in economic activity has led to increasing interest in this issue. The recession caused by the COVID-19 pandemic will eventually eliminate 50 million tourism jobs globally [1], and the impact looks likely to be felt across countries. More importantly, tourism is closely linked to other important industries such as air transportation, oil production, hotels, and retail, so the chain reaction will spread around the world. Scholars, as well as government planners and implementers, must therefore better understand the impact of the pandemic, particularly on tourism.

When assessing the impact of COVID-19, Chinese cities provide a good sample for empirical studies. In China, due to measures such as “city lockdown”, “isolation” and bans on crowd gatherings [2], many Chinese residents have canceled their original travel plans and reduced their willingness to travel in the future, which can be considered a certainty at this stage. In theory, the emergence of crisis events can have a significant impact on tourism and other activities. However, how to quantify the treatment effects from the perspective of the tourism economy in such a short period of time remains an open question to be further explored, and one that is rarely addressed in the existing literature. According to some scholars, based on the experience of SARS, Influenza A H1N1, etc. [16,17], tourism activities may exacerbate the spread of the pandemic, but it will also deal a heavy blow to the tourism industry due to the emergence of the pandemic. Thus, there is a viscous two-way cause-and-effect relationship between the two, which is inherently difficult for us to determine clean treatment effects [18]. But in China’s case, the decisive decisions and enforcement of lockdowns and quarantines, not by a few but almost all of the Chinese people behavior choice, provide us with effective tools to address the endogenous biases between the pandemic and tourism development, allowing this study to more accurately identify the treatment effect of COVID-19. Meanwhile, it can also be found that many variables such as the reduction of residents’ willingness to travel and when they will be able to recover in the future lead to more uncertainty about the future of tourism development than ever before. To sum up, the previous literature mainly (a) analyzes the impact of crisis using post-event annual statistical indicators’ values and trends, and (b) relevant discussions mainly focus on qualitative documents, while empirical research methods are seldom applied, and mostly reflect statistical correlation. It is believed that the main drawback of these studies is that results can be skewed by the lack of real-time index data related to tourism and the neglect of possible endogenous problems. So, that’s exactly what this study is trying to fill in the blanks and provide a comprehensive study of the links between COVID-19 and tourism.

The main contribution of this study is twofold. First, this study crawled the daily Baidu Index of tourism-related keywords by 247 Chinese prefecture-level cities from the website http://index.baidu.com/v2/index.html#/ (accessed on 29 April 2021), and then constructed the “tourism network attention” indicator (TNA) to measure the overall state tourism development in the COVID-19 era in real-time. The use of these data can increase the generalizability of empirical research evidence, as most existing research focuses on the use of annual data or statistical indicators. Secondly, compared with previous research, the empirical analysis has certain advantages in methodology and identification strategy. This study used the regression-discontinuity-design (RDD) method to understand the impact of COVID-19 on tourism. Designed by time cutoff, the study represents pioneering research in pandemic impact assessment. Specifically, the RDD method will be applied to capture the causal link between COVID-19 and the degrowth in tourism. Angrist and Pischke [19] believe that in a highly rules-based world, the “arbitrariness” of some contingencies provides us with good experiments (or quasi-experiments). The RDD, an empirical method second only to randomized trials, effectively uses real-world constraints to analyze causality between variables, avoiding the inherent bias in parameter estimations. Therefore, it can truly reflect the causal relationship between variables [20,21]. In addition, this study also retrieved the Baidu index of some daily life keywords as control variables, and test the continuity of the cutoff position to improve the robustness of the RDD analysis. Principles of relevant methodologies can be found in relevant studies [18,20,22]. What is special and perfect in the methodology is the construction of a differenced RDD regression to further validate the accuracy of identification and obtain a clearer treatment result by subtracting the value of the corresponding day with a one-year lag from the explanatory variable.

The rest of this article is arranged as follows. Firstly, this study introduces the data and identification strategies, briefly introduces the statistical method of the key explanatory variable TNA, and explains the basic details of the RDD model. Then the results are presented and robustness analysis is performed. Finally, the main findings are summarized in the conclusion.

2. Materials and Methods

The basic period covers from 1 July 2019 to 25 April 2020, and to have more reliable results, samples from 1 July 2018 to 25 April 2019, and 1 July 2020 to 25 April 2021 have also been collected, which will be explained later. And the data contains 247 Chinese prefecture-level cities, which covers almost all prefecture-level cities, taking into account the availability of the data. The dependent variable of this study is tourism development. Several annual data indicators from statistical yearbooks are usually used for measurement, such as the number of tourists received, tourism income, etc. [23]. This study focuses on the real-time development of the tourism industry, the annual data is insufficient. Therefore, this study turns to the daily information of TNA, which can be calculated by Function (1) following typical pieces of literature [24,25].

$$TN{A}_{c,t}={\displaystyle \sum _{k=1}^{m}B{I}_{k,c,t}\times w_k}$$

(1)

where TNA_c,t is the comprehensive tourism index of city c on day t, which is constructed based on the Baidu index of tourism-related keywords; BI_k,c,t represents the Baidu index of the k-th keyword for cities, w_k is the corresponding weight which determined under the entropy weight method. Specifically, this study mainly selects six keywords, as “tourism”, “travel agency”, “self-driving tour”, “Ctrip”, “Qunar”, “fun place” in Chinese, to get the BI_k,c,t. The Baidu index provides a weighted sum of the search frequency of certain keywords by network users in all prefecture-level cities of China. It reflects the distribution of information flow of residents searching the relevant keywords on the network, and has the advantage of reflecting the development of tourism in real-time from the demand side, covering a wide range of cities.

In addition, this study tries to add some control variables to the basis of empirical analysis based on existing literature. Unlike the use of annual data, the daily data in the present paper will face the unavailability of global economic and macro indicators. Given this shortcoming, this study makes improvements in the following three aspects. First, the RDD method itself focuses more on information near the cutoff point of the forcing variable. Following that research strategy, this study analyzed the time cutoff of COVID-19 and used local estimations to largely eliminate the interference of macro-environmental changes. Second, using a web-based data crawl on the post-weather report website http://www.tianqihoubao.com/lishi/ (accessed on 29 April 2021), this study collated the weather data of 247 Chinese prefecture-level cities and introduced the daily maximum temperature, minimum temperature, sunny conditions, and wind grade as control variables. Third, a living-related keyword retrieval ratio is constructed based on the Baidu Index, denoted by Live. Theoretically, the outbreak would not affect the search behavior of these keywords, but the focus on these aspects of life could be a factor affecting tourism demand, and therefore be exogenous. In this way, through local restrictions and appropriate direct time effect control, the estimated result is the causal relationship between COVID-19 and tourism that this study is concerned with.

According to the evaluation of tourism development and the related classical literature of the RDD method, the regression model is set up as Function (2).

$$\begin{array}{ll}TN{A}_{c,t}& =\mu +\alpha \times D_{c,t}+{\displaystyle \sum _{i=1}^g{\beta }_i\times f(t)+}{D}_{c,t}\times {\displaystyle \sum _{i=1}^g{\gamma }_i\times f(t)+}\delta \times X_{c,t}\\ & +D{Q}_{c,s}+T{C}_c+u_{c,t}\end{array}$$

(2)

D_c,t is the treatment variable in the RDD model depending on the outbreak date of the COVID-19 event; the identifiable start date is determined according to the process of the COVID-19 crisis in China and the time node that received national attention. Accordingly, it is found that Zhong Nanshan, leader of the high-level expert group of the National Health and Health Commission, academician of the Chinese Academy of Engineering, and expert in respiratory medicine, accepted the CCTV “News 1+1” interview and argued it clearly that the new crown virus is of human-to-human transmission, and that day became the landmark time of COVID-19 and its control, which is the cutoff point of this study. It was 20 January 2020. Therefore, this study set D = 0 before that date and D = 1 after. t is the forcing variable of the RDD, to measure the time distance from the cutoff point, on the cutoff day t = 0. f(t) is a polynomial function of the forcing variable, and RDD will select the optimal order of the polynomial when estimating.

X is the vector of the control variables, including daily maximum temperature, minimum temperature, sunny condition, wind grade, and Live mentioned above. Here, the basic keywords of the variable Live mainly include “food”, “oil price”, “reading” and “cold”. Their Baidu Indexes were crawled and then synthesized into a composite index by the entropy weight method. In addition, the model also controls the regional fixed effect and the fixed effect of the city as the major source of tourists, donated as DQ_c,s, and TC_c respectively; u is the random disturbance item. Descriptive statistics of variables are shown in

Table 1. Variable settings and descriptive statistics.

Variables	Setting	Obs	Mean	Min	Median	Max
Panel A
TNA	Tourism network attention	82,434	58.38	0	39	1297.47
QWmax	Maximum daily temperature	82,434	15.82	−26	17	41
QWmin	Minimum daily temperature	82,434	5.97	−37	7	28
Qing	Whether sunny day	82,434	0.42	0	0	1
Feng	Wind scale	82,434	2.80	1	2	8
Live	Living focus ratio	82,434	42.02	0	30	223.61
Panel B
TNA_diff	Differenced Tourism network attention	81,740	−12.62	−735.23	−8.9	709.42
QWmax_diff	Differenced Maximum daily temperature	81,740	0.68	−28	1	22
QWmin_diff	Differenced Minimum daily temperature	81,740	0.49	−19	1	24
Qing_diff	Differenced Whether sunny day	81,740	0.05	−1	0	1
Feng_diff	Differenced Wind scale	81,740	0.10	−7	0	6
Live_diff	Differenced Living focus ratio	81,740	−19.56	−163.91	−15	58.05

.

Further, this study provided a differenced RDD shown as Function (3). TNA_diff refers to the arithmetic difference between the current period value and the one-year lag period value of TNA. The current period in this study is ranged from 1 July 2019 to 25 April 2020, and the one-year lag period is 1 July 2018 to 25 April 2019, correspondingly, during which time there was no pandemic. And X_diff is the differenced control variables vector constructed in the same way. Function (3) is particularly useful in assessing the impact of COVID-19 on tourism as there is no pandemic in the same period of 2018–2019.

$$\begin{array}{l}TNA\_dif{f}_{c,t}=\mu +\alpha \times D_{c,t}+{\displaystyle \sum _{i=1}^g{\beta }_i\times f(t)+}{D}_{c,t}\times {\displaystyle \sum _{i=1}^g{\gamma }_i\times f(t)+}\delta \times X\_dif{f}_{c,t}\\ +D{Q}_{c,s}+T{C}_c+u_{c,t}\end{array}$$

(3)

It focuses primarily on the coefficient $\widehat{\alpha }$, which is the estimator of the local average treatment effect (LATE) at t = 0. A significant negative result of $\widehat{\alpha }$ would indicate that COVID-19 significantly reduces TNA or TNA_diff, and harms tourism. Besides, by eliminating the samples after the cutoff day by day and carrying out RDD regression one by one, this study can further explore the dynamic nature of the impact of the pandemic. The relative description will take place in Section 3.

Panel A is the corresponding variables obtained from 1 July 2019 to 25 April 2020; Panel B is the differenced result with the one-year lagged period (that is, the same time interval from 1 July 2018 to 25 April 2019), which largely eliminate the impact of unobservable factors. Here this study adds the suffix “_diff” to distinguish them from the basic form. It is especially helpful to evaluate the impact of the pandemic since there is no pandemic in the same period from 2018 to 2019. Therefore, the Panel B data enables this study to apply the differenced model and provide more convincing empirical evidence. At the same time, to further study the recent recovery of China’s tourism market, this study also uses data from 1 July 2020 to 25 April 2021, compared to the same period in 2018–2019 and the same period in 2019–2020 to explore the recovery of China tourism market in the context of effective pandemic prevention and control using quantitative information.

3. Results

In this section, according to Function (2) and Function (3), the results are estimated by applying the shape RDD regression. In particular, to further ensure the accuracy of the results, this study provides three estimators: the conventional RDD estimates, the bias-corrected RDD estimates, and robust bias-corrected RD estimates [18,22]. Then, the robustness of the estimation is tested to observe the validity of the empirical analysis. All estimations are carried out with Stata 16.0.

3.1. Basic Results

The daily distance from 20 January 2020 is taken as the forcing variable to investigate whether the core dependent variables TNA (and TNA_diff) jump or not at the cutoff point. The program written by Cattaneo et al. [26] helps to show the effect of COVID-19 on tourism from data-driven respect, see Figure 1. It shows the results of dependent variables using TNA and TNA_diff under the first-order polynomial and the fourth-order polynomial, corresponding to the upper part and the lower part, respectively. Intuitively, the dependent variables have an obvious jump before and after the cutoff point, and both V and TNA_diff decrease significantly. This conclusion is valid under both first-order and fourth-order polynomials. Therefore, it is clear that the outcome variable in this study has a jump phenomenon at the cutoff point, that is, there is a discontinuity effect, and the jump direction also reflects the adverse impact of the pandemic on tourism development.

To investigate the quantitative results of the impact of the pandemic,

Table 2. Estimations of the impact of COVID-19 on tourism.

	Panel A		Panel B
VARIABLES	TNA	TNA	TNA_diff	TNA_diff
Conventional	−2.40 ***	−2.12 *	−5.26 ***	−10.77 ***
	(0.789)	(1.180)	(0.933)	(2.210)
Bias-corrected	−2.25 ***	−2.47 **	−5.72 ***	−11.98 ***
	(0.789)	(1.180)	(0.933)	(2.210)
Robust	−2.25 **	−2.47 *	−5.72 ***	−11.98 ***
	(0.985)	(1.373)	(1.006)	(2.335)
Observations	82,434	82,434	81,740	81,740
Bandwidth	55.84	39.80	32.86	19.41
TC FE	YES	YES	YES	YES
Region FE	YES	YES	YES	YES
Control Var.	NO	YES	NO	YES

Notes: ***, **and * indicate coefficients are significant at levels 1%, 5%, and 10%, respectively; and standard errors in parentheses.

reports the estimator parameters using the bias-corrected program for RDD. Panel A and Panel B show the estimation results of RDD regression and differenced RDD regression, respectively. Information includes the results without control variables (columns 1 and 3) and the results with control variables (columns 2 and 4). Moreover, both the regional fixed effect and whether it is the major tourist source city fixed effect are controlled in the regression. Panel A shows that the occurrence of the pandemic significantly impaired the development of tourism in the city. The TNA has decreased by 2.40 (p < 0.01) after the outbreak of COVID-19, which is significant at the 1% level, and when control variables are included, it is about the same as the coefficient value −2.12 (p < 0.10) after adding the control variable.

To avoid the bias caused by potential unobservable factors, Panel B reports the results of the differenced RDD regression. Through the first-order difference with the previous year, the model largely eliminates the influence of potential factors, and using the same period can effectively eliminate the interference of seasonal trends. The results indicate that when differenced data is used in the RDD model, the significant impact of the pandemic was further highlighted. The coefficients are −5.26 and −10.77 without and with controls, and both are significant at the 1% level. What’s more, both bias-corrected and the robust coefficients exhibit consistent parameter results, which means the treatment effect of COVID-19 is robust. Therefore, it can be considered that the tourism industry has suffered a significant impact during the pandemic stage, showing that each coefficient is negative and significant.

3.2. Heterogeneity

However, due to the pandemic, the decline in tourism demand in major tourist source cities or regions is likely to be more pronounced, leading to differences between cities. To test this hypothesis, this study uses the sample grouping of whether it is the major source of tourists and then performs the RDD regression analysis grouped. The results are shown in Figure 2 and

Table 3. The impact of COVID-19 on tourism with city heterogeneity.

	Non-Major Tourist Source Cities		Major Tourist Source Cities
VARIABLES	TNA	TNA_diff	TNA	TNA_diff
Conventional	−2.15 ***	−4.57 ***	−14.20 **	−14.91 ***
	(0.746)	(1.112)	(5.683)	(4.382)
Bias-corrected	−2.50 ***	−5.28 ***	−15.81 ***	−17.79 ***
	(0.746)	(1.112)	(5.683)	(4.382)
Robust	−2.50 ***	−5.28 ***	−15.81 **	−17.79 ***
	(0.808)	(1.183)	(7.010)	(5.319)
Observations	70,772	70,176	11,662	11,564
BWselect	mserd	mserd	mserd	mserd
Bandwidth	29.91	20.66	43.99	32.36
Kernel	Triangular	Triangular	Triangular	Triangular
Region FE	YES	YES	YES	YES
Control Var.	YES	YES	YES	YES

Notes: ***, ** and indicate coefficients are significant at levels 1%, 5%, and 10%, respectively; and standard errors in parentheses.

. The upper and lower parts of Figure 2 correspond to the non-major tourist source cities and the major tourist source cities respectively. The scatter and curve positions show that the samples of the two types of cities have a jump in the trend of the dependent variables at the cutoff point, but relatively speaking, the downward jump of the major tourist source cities is larger and more obvious both for TNA and TNA_diff. So, it is necessary to identify the heterogeneity by city grouping regression.

Table 3 shows the heterogeneity of the impact of the pandemic on different types of cities with a subset of samples. For non-major tourist source cities, the coefficient is about −2.15 to −4.57, while the impact coefficient of the major tourist source cities is −14.20 to−14.91, significant at the 1% level, indicating that the sudden outbreak of the pandemic has a greater impact on the major tourist source cities.

3.3. Dynamic Analysis

Over time, the impact of COVID-19 on tourism will change dynamically. To identify this dynamic, this study set the following strategy: sequentially remove samples from t = 1 to t = k (with k = 1,2,3…), reset the driver variables in the new sample range, and conduct RDD analysis, in that way a series of RDD coefficients are estimated, and the trend change will reflect the dynamic impact of COVID-19. In this study, the basic database is updated to the day of 25 April 2021, and the maximum k is set to 365, which means that to see the dynamic effects for one whole year from 20 January 2020. The results are shown in Figure 3, and it shows that effects have a brief recovery period after the cutoff point, then a downward jump, and then show average treatment effects below zero, meaning COVID-19 does harm tourism overall.

Figure 3 intuitively and significantly shows the dynamic nature of the daily moving RDD regression coefficient results, which are characterized by changes over time, reflecting the seasonality of TNA corresponding to the seasonal tourism activities. It is confirmed that the negative effect of COVID is more significant in the tourist season. At the same time, to more clearly express the important holiday effects that scholars are concerned about, this study selects the time interval of two major Chinese holiday weeks in the figure, namely the May Day holiday (International Labor Day) and the October holiday in China (China’s National Day), the two major “golden weeks” of China, here two places showed a significant decrease in TNA.

3.4. Comparison of the TNA States before and after the Pandemic

In this section, this study has collected and calculated the TNA information of each city in China from 1 July 2018 to 25 April 2019, 1 July 2019 to 25 April 2020, and 1 July 2020 to April 2021. The scatter plot and the mean line chart of cities are used for visual presentation of the difference in the trend of TNA at different stages. Here, the 2019–2020 period (pandemic period) is used as the basic period to be compared, and then, the comparison relationships between the 2018–2019 period (the period before the outbreak) and the basic period, and the 2020–2021 period (the period after the outbreak) and the basic period are shown respectively, showing their differences in TNA trends.

To make comparisons more convenient, this study conducted the demeaning processing of individual fixed effects on the TNA data each year. That is, this study can check the changing trend of tourism attention each year under the condition of excluding individual effect differences, see Figure 4. The results show that compared with 2018–2019 and 2019–2020, the 2020–2021 period has shown a more obvious positive upward trend since 2021 when the individual effects of cities are excluded. This result reflects China’s tourism attention and tourism fever has a slight sign of recovery, with strong measures for pandemic prevention and control.

4. Validity and Robustness

4.1. The Validity Test

To ensure validity, the following methods are used. First, check whether the control variable meets the smoothness assumption, that there should be no obvious jump at the breakpoint. Second, check whether there is a jump in the dependent variable at other breakpoints. Third, it is also one of the characteristics of this article. Use TNA18 (the TNA from 1 July 2018 to 25 April 2019) to replace the dependent variable to identify whether there is also an obvious jump. If there is a jump, the treatment effects that have been identified may not be reliable.

The first row of Figure 5 uses Live_diff and WQmax_diff as the placebo outcomes to show the distribution of control variables on the left and right sides of the cutoff point. It can be found that the discontinuity effects of control variables are tiny for the polynomial fit of the control variables are seemingly continuous around t = 0, therefore, indicating that they meet the assumption of smoothness. For the second row, this study uses two placebo cutoff points at t = −20 and t = −34 as the virtual case to test the continuity of TNA_diff, that is three weeks and five weeks before the real cutoff point at t = 0, the RDD plots show that there are no significant jumps around the placebo cutoff points, so it can prove the authenticity of the cutoff point set in this study. The third row shows the results when the previous year’s TNA18 is used as a proxy for the dependent variable in this study. Figure 5 presents the estimation plots under the first-order polynomial and fourth-order polynomial respectively. It is clearly explained that in the lag period without COVID-19, the TNA would not jump at the RDD cutoff, which further illustrates that the inhibitory effect on tourism found in this study is mainly due to the influence of the pandemic. The above analysis shows that the previous analysis using the RDD method is effective and the results are credible. Readers can contact the author for more information.

4.2. The Robustness Checks

In the previous analysis, this study mainly performed the RDD analysis in the form of a four-order polynomial selected by the model. In this section, this study conducts two ways to check the robustness of the pandemic’s impact on tourism. First, this study tested the robustness of the regression results in different polynomial forms. Figure 6 shows the sensitivity of the coefficient $\widehat{\alpha }$ obtained by Function (3) from order 1 to order 6 polynomials. The results show that all of the conventional, bias-corrected, and robust coefficients indicate that COVID-19 harms tourism and they are statistically significant, at least at the 5% level.

Second, the bandwidth length can significantly affect the regression results, and a robust result requires that the bandwidth length be less sensitive. Therefore, this study further manually sets the bandwidth to be 50–200% of the optimal bandwidth, and views the performance of the coefficients, as shown in Figure 7. It can be seen that within the range of 50% to 200% of the optimal bandwidth, the results of the three groups of coefficients remain significantly negative, and maintain the creative significance to a certain extent. This shows that the conclusion is more reliable.

5. Discussion and Implication

5.1. Discussion

In the basic results section, the coefficients indicating the impact of COVID-19 on TNA or TNA_diff have significantly negative values in many different models, which verifies the adverse impact of the COVID-19 pandemic on the tourism network attention. This is an intuitive description of the effects, and it also shows that, to a large extent, TNA is directly related to travel behavior, since when stricter control measures take place, people will spend more time at home. If the focus on tourism network attention were only on entertainment, the index would there have dropped significantly, but it is not.

Heterogeneity analysis shows that cities with different characteristics are affected by the pandemic to different degrees. This is due to the fact that the pandemic has mainly curbed the occurrence of tourism demand, especially in the major tourist source cities, while regions with lower demand tend to remain less willing to travel and are therefore less impacted. The emergence of heterogeneity allows the studies to further identify the potential impact of the pandemic on urban, regional, or other tourism disparities in future studies, and to focus more on the changing and nurturing demands of key source populations.

In the dynamics of the pandemic impacting tourism development, the short-term recovery near t = 0 has greater relevance to the traditional Chinese Spring Festival, but with the development and popularity of the pandemic across regions of China, the damaging effect on tourism development has become more obvious. The overall volatility is relatively high, and significant negative coefficients indicate that in recent months, due to COVID-19 and regional pandemic prevention and control measures, the tourism network attention in China has not shown any obvious signs of recovery. According to the trend of TNA, this study believes that the tourism recovery trend to May Day holiday expected by the media and experts has not come, at least, there is no clear evidence in 2020.

To obtain intuitive discussions on tourism development issues in the context of key holiday effects such as the May Day Holiday and the November Holiday in China, which two are the most important tourist weeks in China in the first half and second half of the year. Figure 3 uses the boxes to highlight the changing trend of the pandemic impact effect based on the differenced RD regressions during a total of three weeks before and after the two holidays. It can be seen that the pandemic does have a significant negative impact on the TNA of cities, implicating that the TNA has a sharp drop before and during the holiday weeks compared to that of the previous year. While June to August is the peak season for China’s tourism market, compared to the same period the previous year, cities have shown a significant decline in the attention paid to tourism online search in the background of the pandemic. The results of the dynamic analysis can intuitively reflect the negative impact of the pandemic on the tourism industry, although China has made sufficient results in the prevention and control of the pandemic. At the same time, more negative and significant coefficients of the impact of the pandemic on tourism are found during important holidays and the tourist peak season. It shows that the impact of the pandemic on tourism is substantial. As the global pandemic continues to spread and continue, the impact of the pandemic on tourism development is not optimistic.

In addition, a slight sign of recovery has been observed in the trend, but it still needs to be understood that the recurrence of the pandemic is one of the determinants of tourism. As the current pandemic situation can be observed, when the pandemic situation becomes serious, the control in local areas will also spread to the tourism development in other areas, which will be reflected first in the changes in the level of TNA. TNA is a reflection of potential tourism demand. Whether the potential demand can be transformed into a real demand also depends on various supports from the external environment. Therefore, it is possible to remain strategically optimistic about the post-pandemic recovery of the tourism market, but it should also be tactically cautious. Grasp the timing of policy support for tourism recovery. Concerning the support and intervention policies for the tourism and residence industry, countries have to find a balance between them to balance the recovery and development of the tourism and residence industry and other aspects of the country.

5.2. Implication

Overall, the severe downturn tourism industry amid the pandemic can be understood on two levels. First, increased economic uncertainty has reduced tourism demand, which is linked to risk aversion in the face of uncertainty. Second, administrative restrictions introduced during the COVID-19 pandemic have the potential to significantly reduce the supply of tourism services, including restricting population mobility. While the pandemic has sparked a tourism crisis in several countries, including China, the tourism industry is likely to be a key sector for the global economic recovery after the pandemic, so it is important to focus on tourism during the pandemic.

This study is of some theoretical and policy implications. Theoretically, strong evidence supporting the damage caused by COVID-19 to tourism development is provided. This study clarifies the dynamics of the pandemic’s treatment effects over time, and there is no obvious recovery process in the overall trend in 2020. By looking at those dynamic impacts, the study sought to capture the likely timing of tourism recovery from recession amid the pandemic. In terms of policy implications, the emergence of the pandemic poses significant economic and political challenges for national and local governments. This study illustrates from the tourism demand side that COVID-19 has greatly impacted residents’ tourism willingness in the short term. As the pandemic is gradually brought under control or recurs, its impact is constantly changing. Therefore, while helping the large-scale development of the tourism industry, the government should make efforts to cultivate and guide tourism demand.

In the post-pandemic era, for the marketing of Chinese tourist cities, suggestions and inspirations are that, on the one hand, the recovery of the tourism market should be identified in time, and tourism demand should be guided and cultivated promptly, so that enterprises and self-employed individuals with the ability to organize tourism demand can quickly operate and demand can converge, and the market can recover. On the other hand, it is necessary to accurately monitor the development of the supply side of the tourism market. COVID-19 has forced many relatively weak tourism businesses to terminate or withdraw, which means that when the tourism market fully recovers, there may be a gap in the provision of tourism services, which should be paid attention to and deployed in advance. As for how to further increase the TNA of Chinese tourist cities, this study finds that tourism consumers have an obvious tendency to “internet celebrity cities” when searching for keywords about tourism. Then take advantage of the current change in the way of information exchange in China’s short video and digital economy era, actively dig out the city’s tourism resources or travel personality, consciously build an Internet celebrity city, increase Internet exposure and media voice, then the city will receive attention.

Given the current global pandemic of COVID-19, China’s timeliness and advancement in responding to the pandemic means that China will enter the post-pandemic era faster and earlier than other countries in the world. Although European and American countries have different measures to combat the pandemic than China, observing the performance and trends of China’s tourism development under COVID-19 can also help provide some potential guidance for other countries, and can also serve as a signal for the recovery of the global tourism industry.

6. Conclusions

COVID-19 has forced countries to take various measures to restrict the movement and gatherings of people. While the pandemic has had a positive impact on slowing the spread of the pandemic, it has also had a huge impact on the economy and society, attracting considerable interest from scholars and policymakers. This study attempts to investigate the treatment effect of COVID-19 on tourism network attention in 247 Chinese prefecture-level cities from 2018 to 2021, to explore the adverse effects and the developing trends of the pandemic on tourism. To this end, this study adopted a very effective RDD method and analyzed it in the form of an intelligent setting. The basic regression results used TNA and TNA_diff as the explained variables, respectively, and the RDD analysis was carried out. The estimated treatment effect coefficient values were −2.12 (p < 0.10) and −10.77 (p < 0.01), which were statistically significant, demonstrating the causal relationship between the pandemic and tourism development in quantitative and empirical terms. And the effects vary with whether it is a major tourist source city or not. Specifically, the estimated coefficient for major tourist source cities is −14.91 (p < 0.01), while that for non-major ones is −4.57 (p < 0.01), that is the impact of the pandemic on tourism is greater in the major tourism source cities. Further identifying the dynamics over time, the impact of the pandemic fluctuates negatively, and this adverse impact is more pronounced during the two major Chinese holiday and tourist seasons. This study further compares the trend changes of TNA in the pre-pandemic period and the pandemic period, and in the post-pandemic period and the pandemic period. The results show that, after excluding the individual effects, the TNA of Chinese cities shows an upward trend in 2021 relative to the year 2020. This is good news for tourism development, but the global spread, mutation, and recurrence of the pandemic remains a key variable to focus on.

Although this study is new to existing literature, it still has its limitations. First, Chinese cities have different economic and tourism characteristics and are at different stages of development. Their heterogeneity made it difficult to make a single recommendation. Spatial and cultural differences must be taken into account. Second, researchers must also pay attention to clarifying and considering theoretical mechanisms for COVID-19 and other contingencies affecting regional tourism development. Third, regarding the rules for dealing with COVID-19, China has stricter rules than others, making the case of China atypical and hard to be referred to other countries directly. Only when these factors are taken into account will the study be able to fully assess the impact of COVID-19.