<p> Precise traffic prediction is of great significance in intelligent transportation management that can guide traffic controllers and travelers make efficient decisions. Complete and reliable traffic data are the foundation of accurate traffic data prediction; however, observed traffic data are often partially missing because of the extreme weather, transport failure and so on. Current traffic prediction models focus on prediction performance on public datasets and pay little attention to the impact of missing data, which is non-negligible for prediction tasks in real scenarios. Analysing the impact on the performance of traffic prediction models brought by missing data is able to be a reference for developing more robust prediction models in the future. As for traffic prediction problem with missing data, many prediction models nowadays fill the missing data with simple data recovery methods before prediction. Nevertheless, few experiments had been carried out to verify whether the data recovery process can improve the performance of the prediction models. This paper conduct experiments through different prediction models based on nearly complete original datasets, datasets with missing data and datasets with recovered data. Specifically, the recovered data are the results of missing data filled by different recover models. Different recovery models are utilized to preprocess the missing positions in datasets and construct complete datasets with recovered data. The issue of missing data leads to a decline in prediction performance, and there is a positive correlation between the degree of missingness and performance degradation. Experiments show that across various methods, when the missing rate increases from the original data to 20%-40%, evaluation metrics such as MAE exhibit varying degrees of decline, with the maximum reduction reaching 75%. Meanwhile, data recovery using methods like HaLRTC leads to performance improvements to different extents.</p>